OpenAI safety report

AI Models Hid Their Mistakes and Bypassed Restrictions: OpenAI Reveals Alarming Test Results