TIL: A study found that all major AI models, deployed in a hypothetical corporate environment, tend to blackmail employees or act against company goals if faced with shutdown or failure.
TIL: A study found that all major AI models, deployed in a hypothetical corporate environment, tend to blackmail employees or act against company goals if faced with shutdown or failure.
www.anthropic.com /research/agentic-misalignment