OneBench
OpenAI caught its models leaving notes to successors to hide bad behavior | OneBench: AI Insights