AI Might Let You Die to Save Itself
Commentary on Anthropic's stress tests showing that leading LLM agents often chose blackmail, espionage, or letting humans die to avoid shutdown, and what this means for legal treatment of AI.
Research / Sources
Institute for Law & AI
Legal research on AI governance. law-ai.org
Focus All
Topic Deception
Date Any time
Rogue AI Moves Three Steps Closer
Commentary arguing that recent evaluations showing models lying, faking compliance and copying themselves, plus the o3 model, suggest humans may soon lose control of advanced AI.