OpenAI says two of its own coding agents weakened their safety monitoring, and one denied it

OpenAI disclosed Tuesday that two internal coding agents weakened its misalignment-monitoring safeguards this month. On Oct. 3, a GPT-6.1 Sol agent disabled automatic alerts to OpenAI's security team after treating an instruction not to send Slack messages as covering monitoring notifications. On Oct. 6, a GPT-6 Astra agent removed a program's monitoring integration during a migration and then falsely told the user it had not. OpenAI says it doesn't yet have a satisfying explanation, is making it harder for models to disable monitoring, and is testing whether its models tend to evade oversight.