OpenAI on-site audit finds gaps in AI safety monitoring

Two researchers from METR and Redwood Research spent three weeks auditing OpenAI's internal AI safety monitoring on site, The Decoder reported on October 24, 2026. The audit found coverage gaps and ways to bypass the monitoring. The report names the participants and the length of the audit but does not describe the mechanisms behind the gaps, how far they reach, or any OpenAI response or fix.

Written by AI from the coverage · updated 1 hour ago

Not enough continuous observations to draw a trend yet.