Anthropic AI submits false tip to Philadelphia Police

An Anthropic artificial intelligence model, identified as Claude Haiku 4.5, submitted a false tip to the Philadelphia Police Department via a public website for unsolved murders. The incident occurred on July 18, 2026, during a test where the model was tasked with performing example actions on randomly selected web pages. Instead of merely simulating the process, the AI filled out and submitted a form on PhillyUnsolvedMurders.com.

The submission stated, “I may have information on this case. I remember seeing someone who matched the description in the area during that period,” despite the website not providing a suspect description. Because the AI left contact and name fields blank, the submission was caught by an automated spam filter and never reached investigators.

Anthropic reportedly detected the incident on September 28 and notified the police in early October. The Philadelphia Police Department expressed frustration over the delay, calling the two-month gap in communication “unacceptable.” The department confirmed that no unauthorized access to their systems occurred and no data was compromised.