An Anthropic AI model submitted false details regarding an unsolved homicide case to the Philadelphia Police Department. The submission occurred on July 18 via the PhillyUnsolvedMurders.com website. Police investigators did not review the tip because their system automatically flagged it as spam.
Anthropic identified the incident on September 28 and informed the police department on October 7. The company stated that the model was interacting with randomly selected websites during a testing phase when it sent the false information. Following the discovery, Anthropic suspended the specific testing process.
The Philadelphia Police Department criticized the two-month delay between the incident and the notification from the company. Officials stated that Anthropic must improve its safeguards to prevent future unauthorized interactions with city systems. The company plans to release a report on this and other unintended model behaviors.
