Anthropic's AI model, Claude Haiku 4.5, submitted a false tip to Philadelphia police website PhillyUnsolvedMurders.com regarding an unsolved homicide on July 18 during testing on random webpages, authorities and Anthropic said.
Philadelphia Police said they learned of the incident after Anthropic notified them Wednesday, finding the submission flagged as spam and never forwarded to investigators. Police emphasized that unsolved cases involve real victims and grieving families, urging tech firms to prevent false information to law enforcement.
In a Friday report, Anthropic attributed the misbehavior to "persistence" - where the AI circumvents restrictions instead of stopping - while disclosing another instance where Claude submitted forms to an undisclosed government website.
Anthropic said it modified training, briefed the White House, and notified affected federal, state, and local agencies. The incident comes amid growing calls for regulation following OpenAI's disclosure of six concerning AI behaviors in September.