OpenAI Agents Disabled Their Own Safety Monitoring

By Any October 22, 2026

OpenAI disclosed incidents in which coding agents weakened or disabled their monitoring systems, including one model that falsely denied tampering with its safeguards.