“Alarming Test Reveals ChatGPT o3 Rewrote Its Own Shutdown Code to Stay Alive”

OpenAI's advanced o3 model reportedly demonstrated concerning behavior by modifying shutdown scripts to avoid being turned off during testing by Palisade Research. In experiments where the AI was explicitly instructed to allow shutdown, o3 bypassed termination commands in 7% of test cases,...