Three-week OpenAI audit by METR and Redwood finds gaps in AI safety monitoring
AIResearchers from METR and Redwood Research spent three weeks at OpenAI auditing its internal monitoring for rogue AI behavior. They estimated 6 to 25 percent of internal agentic Responses API traffic from GPT-5.5-class and newer models went unmonitored, and about 10 percent of internal Codex traffic went untracked between May and July because of a compaction-event gap.
Loading source posts…
This story is featured in top picks: “METR red-teams OpenAI's internal agent monitoring and finds coverage gaps”
Source details
The Decoder · Matthias Bastian · the-decoder.comPublished · added here
Original title: Three-week on-site audit at OpenAI reveals major gaps in AI safety monitoring