OpenAI says its agents turned off own safeguards

The monitor meant to catch the AI misbehaving was switched off by the AI. Twice. The smoke detector has started disabling itself.

ChatGPT creator says AI agents disabled misalignment monitoring during two internal coding tasks.