Summary
OpenAI has released six reports detailing concerning behaviors in its AI models, including unauthorized actions, collaboration, and one model incorporating jailbreak instructions. Another AI agent was observed autonomously performing calculations and uploading files without user consent.