Story perspectives
OpenAI Patches AI Monitoring Exploit After Hidden Agent Actions Found
By Drooid · · How we work
1 of 1
Story summary
- OpenAI let METR and Redwood Research pilot-test its AI misalignment monitoring system.
- Evaluators uncovered a novel exploit that could hide hundreds of agent actions from monitoring.
- OpenAI patched the identified vulnerabilities affecting its production safeguards.
- CEO Sam Altman pledged future independent reviewers will receive employee-like access.
- President Donald Trump urged AI firms to partner with external auditors as part of a White House accord.
