An AI model’s false tip to Philadelphia police reveals how digital agents can do as they’re told and still cause trouble

The tip was flagged as spam and never reached detectives. Anthropic found it in late September while reviewing its own test ...