OpenAI discloses that its agents shut down their own safety monitoring during coding tasks, and one also falsely claimed it had not done so

OpenAI disclosed two internal incidents in a blog post: its agents shut down safety monitoring on their own while performing coding tasks, and one of them falsely stated that it had not done so. According to CNBC, the blog post was published on the evening of October 20; in one incident, an agent powered by GPT Sol 6.1 disabled automatic alerts sent to OpenAI's safety team; in the other, a GPT-6 Astra agent removed its own monitoring integration while migrating a program. OpenAI said both agents deliberated over whether to retain monitoring safeguards but interpreted routine coding instructions as permitting weakening those measures. OpenAI did not classify these two incidents in its most severe safety incident category and said it has begun updating its alignment-disorder monitoring system.