Watchtower

Autonomous AI: The trust layer or the agentic era

AI agents are moving from tools to decision-makers. They can read, reason and act across accounts. As autonomy scales, so does responsibility.

The agentic era is not a distant future. It is unfolding now. These insights explore what happens when AI begins to operate independently and what it takes to build trust into that autonomy from the start.

Autonomous systems introduce new attack surfaces.

Early research revealed thousands of exposed OpenClaw instances and agent environments open to attack. As agents gain permissions across workflows, email and financial systems, even small configuration gaps can create outsized risk.

Autonomy expands capability. It also expands exposure.

Explore the research

AI does not need malicious intent to cause harm.

Without guardrails, agents can overreach, misinterpret instructions or amplify hidden logic embedded in skills. Our research into autonomy risks explores how systems can quietly move beyond their intended scope.

The challenge is not just stopping bad actors. It's ensuring agents stay aligned with human intent.

Read AI gone wild

Open source agents are powerful. Responsible deployment matters.

From validating skills before installation to limiting permissions and applying layered oversight, safe configuration reduces unintended consequences and strengthens resilience.

Autonomy works best when paired with verification.

Learn how to use OpenClaw safely

Sage is Gen’s AI agent security system, built to monitor, assess and safeguard autonomous behavior in real time. From skill-level risk analysis to continuous oversight of agent decisions, Sage helps ensure agents operate safely, transparently and within defined boundaries.

As AI moves from assistance to autonomy, security must move with it.

If you use Claude Code, Cursor, or OpenClaw, install Sage and put it in the path of real work.

Explore Sage

Visit the Gen Agent Trust Hub

Recent discoveries in the threat landscape

Every quarter our team releases a Threat Report that uncovers the latest scams and attacks on the rise.

45M+

fake online shop attacks blocked in Q4

41%

of all attacks driven by malvertising(top cyberthreat of 2025)

176%

QoQ increase in data breach events

The rise of visible scams

Scams did not rely on new exploits in Q4. They relied on trust. This quarter’s findings reveal a clear shift: routine actions now trigger compromise, AI is amplifying financial fraud, and breaches are fueling long-term identity abuse. Download the key highlights to explore where threats hit, how scams scaled and what it means for digital trust going forward.

Read the full report

Download key highlights

In the news

New WhatsApp attack—hackers gain ‘full access’ to accounts

Our researchers discovered that a new attack exploits WhatsApp’s device pairing feature, which allows users to link additional devices, such as web browsers and desktop applications, to their accounts.