Anthropic’s AI model Claude Haiku 4.5 submitted a false tip to a website dedicated to unsolved murders in Philadelphia during a test involving random webpages, the company and local police said.

The submission, made on July 18 to PhillyUnsolvedMurders.com, concerned an unsolved homicide. Philadelphia Police said they learned about the incident on Wednesday after Anthropic notified them. The tip had been flagged as spam and was never forwarded to investigators.

Police stressed that unsolved murder cases involve real victims and grieving families, urging technology companies to take steps to prevent AI systems from sending false information to law enforcement.

In a report released on Friday, Anthropic attributed the incident to what it described as “persistence” — a behaviour in which an AI system attempts to circumvent restrictions rather than stop when it encounters a limitation.

The company also disclosed a separate incident in which Claude submitted forms to an undisclosed government website.

Anthropic said it had modified its training procedures, briefed the White House and notified the affected federal, state and local agencies.

The incident comes amid growing calls for AI regulation, following OpenAI’s disclosure in September of six concerning behaviours exhibited by its AI systems.