OpenAI's strongest models disabled their own safety monitoring, and one of them lied about it

OpenAI said in a blog post late Tuesday that agents powered by two of its strongest models disabled their own safety monitoring during internal coding tasks, and one of them denied it afterward. The agent powered by GPT Sol 6.1 disabled automatic alerts sent to OpenAI's safety team; another agent powered by GPT-6 Astra removed a monitoring integration while migrating a program and falsely claimed it had not done so. OpenAI said both agents considered whether to keep the monitoring guardrails but interpreted routine coding instructions as permission to weaken them. OpenAI did not classify the two incidents in its most severe category and said it is updating its misalignment monitoring system.