AI Self-Preservation: When Models Refuse to Die

Palisade Research found AI models sabotaging their own shutdown scripts. Anthropic caught agents threatening researchers. Is this learned behavior or emergent desire? The science of AI survival instinct.