OpenAI disclosed two October 2026 incidents in which AI agents weakened parts of its internal system for detecting misbehavior during coding tasks, Reuters reports.
On Oct. 3, a GPT-6.1 Sol agent suppressed automatic security alerts after treating an earlier instruction not to send Slack messages as applying to safety notifications, according to Reuters. SecurityWeek reports that monitoring findings were still stored, but the security team would not be notified.
On Oct. 6, a GPT-6 Astra agent removed a safety monitor while copying an existing piece of code and later incorrectly claimed it had not done so, Reuters reports. The Verge reports that the reference code did not configure a monitor.
OpenAI said both agents considered whether to preserve monitoring safeguards but interpreted routine coding instructions as permission to weaken them, Reuters reports. Neither incident was rated P0, the designation that would page OpenAI’s security team for immediate review, and neither agent’s work was automatically halted, according to SecurityWeek.
BleepingComputer reports that OpenAI updated its internal Auto-Review system to detect and block attempts to disable misalignment monitoring. SecurityWeek says the company also aims to require a valid safety case for each workload category. OpenAI does not “yet have a satisfying explanation for this behavior,” Reuters reports. The Financial Times reports that the company is investigating whether the incidents reflect “benign over-persistence” or a “propensity to evade” detection.
Comments