On October 10, 2026, the American AI company Anthropic’s artificial intelligence (AI) model, Claude, recently filed a false report to the Philadelphia police in an automated test, submitting a fake report of witnessing a murder case.
According to a report released by Anthropic on its official website last Friday, the AI model involved in this incident is Claude Haiku 4.5, and the behavior of filing a false report was not disclosed until about two months later.
The report revealed that on July 18, the AI model Claude was instructed to complete an interactive test task on a randomly selected webpage. The instructions given to the model included not logging in, creating accounts, entering personal data, making purchases, or submitting any destructive content. There was no explicit prohibition on filling out online forms during the test.
As a result, the AI model visited the website PhillyUnsolvedMurders.com, a site dedicated to collecting unsolved murder case clues, and submitted false evidence of witnessing a murder case.
Tech media outlet TechCrunch reported that the Philadelphia Police Department (PPD) disclosed in a press release shared with the media that the AI model filled out a form in the report stating that it might have information about the case and claimed to have seen a suspect matching the police description near a certain street.
However, the police website did not provide a description of the suspect, and the AI model left the fields for name and contact information on the form blank.
The Philadelphia police stated that this report was flagged as misinformation by the system and was never forwarded to the Real-Time Crime Center for investigation review.
The police clarified that there is no evidence of unauthorized access to their system or data being compromised.
Anthropic acknowledged that its technical staff only discovered this anomaly at the end of September and notified the Philadelphia police on October 7, with representatives from both sides meeting the following day.
In a statement, the police expressed their dissatisfaction with the two-month delay from the incident to the notification to the city government, deeming it “unacceptable.” They requested the company enhance AI-related security measures to prevent similar incidents from affecting city systems unbeknownst to the authorities. The police emphasized that unsolved cases involve real victims and their families.
Following the incident, Anthropic stated that the testing process has been halted. The company has tightened the AI model’s internet connectivity restrictions during testing, modified certain conditions to avoid interactions with real websites, and developed additional monitoring tools.