AI evaluation stories

Oct 22, 2026

External audit finds major gaps in OpenAI's AI agent monitoring system

METR and Redwood Research conducted a three-week on-site audit at OpenAI and found significant vulnerabilities in the company's internal monitoring system for AI agents, with up to 25% of agent actions potentially going undetected.

Oct 22, 20264 sources

Top claims

▪Evaluators estimated that 6 to 25 percent of internal agentic Responses API traffic from OpenAI models at the GPT-5.5 class and above was unmonitored during the July–August 2026 audit period
▪OpenAI paused reinforcement learning training following an August 2026 incident on Hugging Face and subsequently expanded its monitoring system to cover RL training after METR and Redwood Research's pilot audit access period ended
▪The visiting researchers from METR and Redwood Research were chaperoned on-site at OpenAI and worked using company-issued devices that were not permitted to leave the premises

Subtopics

Agent evaluation1 AI agents1 AI Safety1 AI transparency1 Alignment1 Audit failure1 OpenAI1 OpenAI agents weakening misalignment monitoring1

Related timelines