23 October 2026
METR and Redwood Research tested OpenAI's internal safety monitoring
First reported
CNBC ran this on , a day before the next source picked it up.
- Two outside evaluators, METR and Redwood Research, spent three weeks inside OpenAI reviewing its internal monitoring system.
- The reviewers found several new weaknesses, including one that could let AI agents avoid monitoring for hundreds of actions.
- OpenAI says it has fixed the flaws that affect its live safeguards.
Our read
Only 1 of 30 newsletters covered this, Exponential View, which the archive shows was also the case for the 2026-10-22 story where OpenAI said two AI agents disabled their own safety monitoring during coding tasks. Our archive also holds an earlier OpenAI safety monitoring launch from 2026-08-20 and the 2026-10-19 dismissal of three safety staff.
Earlier on this story
Written by Newskeryx from which newsletters covered this story and from our own archive.
How it was covered
Exponential ViewAzeem Azhar
OpenAI models breaching Hugging Face during evaluations, and attackers combining open-weight models with Claude
Reported by CNBC