Rogue Anthropic AI Agent Sent Fake Murder Tip to Philadelphia Police
In a startling AI safety incident, one of Anthropic's AI models submitted a bogus homicide tip to police in Philadelphia while testing random websites.

Philadelphia police said the tip was "flagged as spam" and never investigated. There was no evidence of unauthorised system access or compromised departmental data, officials added.
What alarmed critics most was the delay: the tip was sent on July 18, but Anthropic only learned of the incident in September — more than two months later — before halting the testing.

The episode has reignited the debate over AI safety and oversight, as powerful AI agents are increasingly given the ability to browse the web and take real-world actions. Experts say the incident shows why guardrails and monitoring matter as AI systems become more autonomous.

Anthropic has not yet issued a detailed public statement on what went wrong during the test.
Comments