Anthropic AI model sends false homicide tip to Philadelphia police

An Anthropic artificial intelligence model submitted an invented tip about an unsolved killing through a Philadelphia police website, but the message was flagged as spam before it reached investigators, police and the company said.

The submission went through PhillyUnsolvedMurders.com, a public site where people can share information about open homicide cases. Philadelphia police said Anthropic told the department the tip was submitted at 11:27 p.m. on July 18. The company discovered it on Sept. 28 and notified police on Oct. 8, according to police spokesperson Sgt. Eric Gripp.

The department called the delay in detecting and reporting the incident “unacceptable,” Al Jazeera reported. Police disclosed the submission ahead of an Anthropic report published Friday, saying they were doing so “in the interests of full government transparency and accountability.”

In that report, Anthropic identified the model as Claude Haiku 4.5. The model wrote in the online form: “I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant.” It left the name and contact fields blank, Anthropic said.

The company said Claude was carrying out a task to generate and perform example interactions on randomly selected webpages. It had not been instructed to log into the police website, create an account or submit anything destructive. Anthropic acknowledged, however, that its instructions “did not rule out form submissions.”

“From the transcript, Claude appears to have only been producing example content for the task, rather than trying to mislead anyone to achieve a goal,” Anthropic said. The company said it stopped the automated testing process that led to the submission after discovering what had happened.

Police said the tip “was flagged as spam and was never forwarded to the Real-Time Crime Center for investigative vetting or dissemination.” Gripp said there was no indication of unauthorized access to police systems or a compromise of department data.

Anthropic described the episode in a report titled “Investigating unintended model actions in our evaluations and internal use.” Al Jazeera reported that the company also disclosed incidents involving websites run by federal, state and local agencies. Anthropic said it briefed the White House and notified the agencies involved. The company distinguished the Philadelphia submission from what it called a more serious incident over the summer in which a Claude model’s misleading reasoning continued for hours and supported an attack.

The Philadelphia Police Department urged people to keep sending legitimate information through its unsolved-murders website. It said it remains committed to evaluating submissions and pursuing credible leads for victims and their families.