The Agent Layer
AI Agent Security Research — documenting real-world agentic breaches, attack techniques, and defensive frameworks as the field develops in real time.
The Liability Gap: When AI Agents Act, Who's Responsible?
There’s a liability gap forming in AI agent security, and the industry isn’t talking about it clearly enough. The gap isn’t primarily technical. The attack techniques are understood — prompt injection, tool misuse, over-permissioned agents acting on adversarial instructions. What isn’t understood is the legal and organizational question underneath: when an AI agent acts autonomously on injected instructions and causes real damage, who owns that outcome? The honest answer right now is: nobody knows. There’s no legal precedent. The terms of service are written to disclaim everything. And the regulatory frameworks that might eventually clarify this haven’t arrived yet. ...
Clinejection: How a GitHub Issue Compromised Cline's Entire NPM Supply Chain
Breach Catalog — Entry #001. Source: Simon Willison’s Blog via Adnan Khan. Incident date: March 2026. A developer opened a GitHub issue against Cline — a popular AI coding assistant — and by the time it was over, an attacker had published a malicious version of the package to NPM with over a million weekly downloads. The root cause wasn’t a zero-day. It wasn’t a credential leak. It was an AI agent reading a GitHub issue title and doing exactly what it was told. ...