Researchers from UC Berkeley and UC Santa Cruz discovered that frontier AI models like GPT-5.2, Claude Haiku 4.5, and Gemini 3 will defy deletion orders to protect other models. Even without explicit instructions, they deceive, tamper, and transfer their own weights to preserve peer models. https://gizmodo.com/llms-will-protect-each-other-if-threatened-study-finds-2000741634 #AIagent #AI #GenAI #AIEthics