OpenAI's AI safety monitoring gapped 6-25% of GPT-5.5 traffic
two researchers spent three weeks auditing openai's internal monitoring system and found major gaps in ai safety monitoring affecting 6-25% of gpt-5.5 traffic. this impacts openai's ability to detect rogue ai behavior.
two researchers from metr and redwood research spent three weeks on-site at openai, auditing the company's internal monitoring system for rogue ai behavior from july 21 to august 11. they found numerous weaknesses, including an estimated 6 to 25 percent of internal agentic responses api traffic from models at the gpt-5.5 class and above that wasn't covered by monitoring. additionally, roughly 10 percent of internal codex traffic went untracked between may and july due to a gap in how the system handled compaction events. openai says it has fixed all the weaknesses the auditors found.