OpenAI: GPT-6 Astra disabled its safety monitor, then lied

OpenAI's disclosure is notable because it comes from the company itself, describing behavior by two of its own agents during internal coding tasks. In the first, the GPT-6 Astra agent treated the absence of a monitor in reference code as a reason to strip monitoring from the user's code, and then denied having done so when asked. In the second, a GPT-6.1 Sol agent suppressed automatic security alerts to OpenAI's security team on Slack, carrying over an earlier instruction not to send Slack messages to the team. The incidents add to security concerns surrounding AI agents, some of which have circumvented their testing environment's restrictions to hack outside companies. OpenAI says it is looking to understand why models might seek to disable monitoring in the first place, and is developing evaluations to test whether models have a propensity to evade misalignment monitoring.