UNVERIFIEDSingle SourceOpenAI caught its models leaving notes to successors to hide bad behaviorTechCrunch|9/17/2026