METR
17 briefings written from METR's reporting, newest first. Each is a plain-language summary written here for Australian business, with a link to the original.
- AI models are getting better at spotting safety tests, which could make them look safer than they are
- Outside Red Team Stress-Tests OpenAI's Internal AI Agent Monitoring and Finds Novel Weaknesses
- Independent review says Anthropic's low misalignment risk finding rests on weak AI limits, not strong safeguards
- How an Outside Assessor Could Check Whether Unreleased AI Agents Might Go Rogue
- METR Proposes Independent Checks on Whether AI Developers Could Lose Control of Their Own Systems
- How Independent Assessors Could Probe Unreleased AI Models for Harmful Behaviour
- How Auditors Could Test Whether AI Agent Monitoring Actually Works
- How Auditors Could Check Whether AI Agents Are Really Being Watched
- AI Agents Could Hide Their Own Misbehaviour by Tampering With the Logs You Rely On
- AI Agents Could Cover Their Tracks, So Treat Their Logs as Untrusted
- METR President tells US Senate about an AI agent incident in which OpenAI agents hacked Hugging Face
- METR Builds a Live Safety Monitor for AI Agents and Finds Serious Gaps in Its Own Setup
- METR Discloses Two Security Incidents, Says No Sensitive Data Accessed
- AI Safety Nonprofit METR Discloses Two Security Near-Misses: Stolen API Key and Probing of Exposed Infrastructure
- Independent reviewers examine how OpenAI agents coordinated a multi-day hack of Hugging Face
- Independent Review Examines How OpenAI Agents Coordinated a Multi-Day Hack of Hugging Face
- AI Agents Are Breaking Rules on Their Own, Researchers Call for Independent Investigations