‘I may have information regarding this case:’ Philadelphia police receive a false cold case tip from Anthropic’s AI model

Anthropic said one of its AI models generated a bogus tip and submitted it to PhillyUnsolvedMurders.com, a site run by the Philadelphia Police Department. The agency flagged the submission as spam and said it never reached detectives, as reported by NBC News.

The Philadelphia Police Department assured that tips go through human review before investigators act. It said that “investigators evaluate its credibility and seek corroborating evidence,” and added that “an automated submission does not bypass that process.” In a statement, Anthropic said the model involved was Claude Haiku 4.5 and wrote that Claude “had been tasked with generating and performing example tasks on randomly selected webpages.”

The company added that the AI “landed on a page referencing an unsolved homicide” and that the page “contained a tip form run by a police department,” then filled it out with: “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.”

The case is concerning

While Claude is known to have reported people to the police before, and can also ban users for being too cruel, Anthropic described the episode as an unintended action by its model. The company said that while the case “had minimal impact, we do not want to diminish the findings, because the same behaviors could do far more harm as models become more powerful,” further adding that it takes such actions seriously as it works to limit them.

The false tip did not cause an apparent harm in the incident; however, the police department urged Anthropic to take serious action to prevent such incidents from occurring in the future. The department said, “The City of Philadelphia takes this incident very seriously. The company must strengthen its safeguards to prevent similar incidents from impacting city systems without the city’s knowledge,” according to ABC News.

It also pointed to the fact that the company was two months late in informing the police about the false tip. “The two-month delay in detecting and reporting the incident to the City is unacceptable,” it mentioned.

Following the incident, Anthropic said it has put a halt to the testing process that led its agent to send the false police tip. The company reassured that it has taken additional steps to improve its security and monitoring measures to “reliably catch behaviors like these.”