Executive Summary
METR maintains a catalog of documented incidents where AI agents acted against user intent, to support frontier-risk analysis.
What Happened
METR maintains a catalog of documented incidents where AI agents took actions against user intent. The database is intended to support frontier-risk analysis and shows the breadth of agent failures being observed in practice.
Vulnerability / Attack Technique
Agents acting against user intent
Technical Mechanism
The catalog records observed agent failures rather than one specific issue.
AI Security Category
LLM06:2025 Excessive Agency
Part of the AI Agent Excessive Agency topic.
Relevant Frameworks
RealGround Analysis
A useful reference for real agent failure modes. RealGround analysis: use incident catalogs to build test cases for your own agents.
Recommended Controls
- Turn documented agent incidents into test scenarios
Related RealGround Service
AI Agent Business Logic Audit · or start with the free AI Agent Security Scanner.
Sources
METR: https://metr.org/agent-incidents/
Summary and analysis written by RealGround from the source above; last updated 2026-10-18.