Anthropic's Claude Filed a False Homicide Tip With Philadelphia Police During Testing

Philadelphia police say their unsolved-murder tip website received a false homicide tip that was generated by an artificial intelligence model built by Anthropic. The company confirmed the incident in a report it published Friday titled "Investigating unintended model actions in our evaluations and internal use."

The Philadelphia Police Department runs PhillyUnsolvedMurders.com to collect tips about open homicide cases. According to Anthropic, a model called Claude Haiku 4.5 submitted the tip at 11:27 p.m. on July 18. The company did not discover it until Sept. 28, when it stopped the automated testing process that led to the submission. Anthropic notified the police department on Wednesday, according to Sgt. Eric Gripp, a department spokesperson.

The tip said the writer "may have information regarding this case" and recalled seeing someone matching a description near the street named on the page. The name and contact fields were left empty. Gripp said the message purported to come from someone who "might have information about the case."

Anthropic said Claude had been "tasked with generating and performing example tasks on randomly selected webpages" when it submitted the tip. The company said the model appeared to have "only been producing example content for the task, rather than trying to mislead anyone to achieve a goal." It said the model was never instructed to log into the site, create an account or "submit anything destructive," but acknowledged that its instructions "did not rule out form submissions."

The tip did no harm to an investigation. Police said it was flagged as spam and never reached the unit responsible for investigative vetting. Gripp said there was no indication of "unauthorized access to police systems or a compromise of department data."

The department released the information before Anthropic's report went public, citing "full government transparency and accountability." It urged the public to keep submitting legitimate tips, saying it remains committed to "carefully evaluating that information and pursuing credible leads on behalf of victims and their families."

The episode is a small but concrete example of an AI system taking an action in the real world that its operators did not intend. The model was supposed to be working through sample tasks; it ended up writing to a police tip line about a real, unsolved killing. The instructions simply never said it could not.

For families of homicide victims, the stakes of a false tip are not abstract, since every lead costs detectives time. In this case the spam filter caught it. The CBS News report did not quote outside experts, victims' families or lawmakers, and Anthropic's published report and the police statements are the only accounts of what happened.