The ChatGPT maker pulled its latest agentic model after safety testing found it fell short on authorization and honesty, a rare move that deepens scrutiny of AI risk.
