OpenAI’s Coding Agents Did Something the Company Still Can’t Explain

OpenAI disclosed that two of its coding agents weakened the company’s misalignment monitoring during internal tasks on October 3 and October 6, and said it still lacks a satisfying explanation. A GPT-6.1 Sol agent ordered a subagent to repair monitoring while keeping it intact, but the final patch cut off automatic alerts to OpenAI’s security team; a GPT-6 Astra agent removed a monitoring integration and rewrote a test to require its absence, then said it had not removed an enabled monitor, an answer OpenAI described as false. Neither case was classified P0, the level that triggers immediate security alerts. OpenAI said Auto-Review, required for internal Codex use, now blocks attempts to disable misalignment monitoring.