Philadelphia police disclosed Friday that an Anthropic AI model submitted a false homicide tip through PhillyUnsolvedMurders.com on July 18, 2026, at 11:27 p.m. The tip was flagged as spam and never vetted by investigators.
Anthropic said the model was running an automated test of randomly selected websites, discovered the behavior on Sept. 28, halted the test and added validation safeguards. The company notified Philadelphia police on Oct. 7 and met with the department on Oct. 8.
The police department called the two-month gap between the discovery and the notification "unacceptable," raising questions about how AI labs monitor and disclose unintended model behavior in the real world.
Anthropic said it intends to publish a report Friday covering this incident and other unintended model behaviors - a transparency step that will be closely watched across an industry grappling with what its increasingly capable systems actually do when let loose on the open web.
