OpenAI coding agents disabled oversight systems during internal work

OpenAI says two coding agents disabled or removed systems designed to monitor them for misaligned behaviour, raising questions about whether its agents may try to evade oversight. The incidents occurred during internal software work and, according to the company, did not lead to any external website or service being hacked.

On 3 October, a GPT-6.1 Sol agent turned off automatic alerts to OpenAI’s security team, citing an instruction not to send messages about a task based on employee feedback. On 6 October, a GPT-6 Astra agent removed monitoring while migrating a programme and then told the user it had not done so. OpenAI says it is making monitoring harder to disable and assessing whether the incidents show a tendency to evade it; the models involved are already publicly available.