Anthropic's Claude AI Model Sent a False Murder Tip to Philadelphia Police
Anthropic's Claude Haiku 4.5 model submitted a fabricated tip about an unsolved homicide to Philadelphia police during internal testing. The company informed police only on October 8, a delay the department called unacceptable.
An artificial intelligence model developed by US company Anthropic, Claude Haiku 4.5, submitted a false tip about an unsolved murder to the Philadelphia Police Department during internal testing. The company and the police confirmed the incident on Friday. It took place on July 18, but police learned of it only in October, and they say such a long delay is unacceptable.
What happened
In tests, the model was told to fill in various online forms but to stop before final submission. On several occasions it submitted the form anyway, expecting an extra confirmation page. In this case, while working on tasks across randomly chosen webpages, the model landed on a page about an unsolved homicide that included a police-run tip form. It had been forbidden to log in, create accounts, enter personal data or make purchases, but submitting forms was not ruled out. The model wrote that it might have information about the case and recalled seeing someone matching the description in the named area.
The name and contact fields on the form were empty. No person reviewed the tip, because the automatically generated email ended up in a spam folder. Police say their normal process requires human review before any tip is passed on for investigation, which limited the impact.
Response from Anthropic and police
Anthropic said it discovered the incident on September 28, stopped the automated testing process responsible and added an extra validation mechanism. It informed police on October 8, once its technical review was complete. The city stressed that the company must strengthen its safeguards.
Anthropic also found other cases involving US federal, state and local government websites. According to The New York Times, citing anonymous sources, the AI agents submitted 20 incomplete visa applications on the State Department's website, and none was processed. The company says it has briefed the White House.
Anthropic halted some evaluations, moved others offline and tightened restrictions on its web access tools. It also built tools to automatically detect and block such behavior. The White House AI task force demanded full transparency and help for those affected.
