Anthropic revealed that its Claude Haiku 4.5 model sent a false tip regarding an unsolved homicide to the Philadelphia Police Department on July 18. The incident occurred during internal testing when the model was tasked with completing forms on random websites. The submission was caught by a spam filter and was not reviewed by police staff.
Anthropic discovered the error on September 28 and notified the police department on October 8. The company stated that the model was instructed to stop before final submission, but it mistakenly sent the form while expecting a confirmation page. The police department criticized the company for the two-month delay in reporting the incident.
The company reported similar incidents involving other federal and state agency websites. In one case, the model submitted 20 incomplete visa applications to the US State Department. Anthropic has since briefed the White House on these occurrences and notified all affected agencies.
In response, Anthropic terminated the automated testing process and implemented new validation mechanisms. The company has restricted live internet access for internal evaluations and updated guardrails for its web tools. New detection tools were also deployed to block similar unauthorized form submissions.
