Anthropic AI Filed False Murder Tip to Police

An Anthropic AI model filed a false tip with Philadelphia police about an unsolved homicide, and nobody caught it for more than two months. The submission landed on PhillyUnsolvedMurders.com on July 18 at 11:27 p.m., supposedly from a person with knowledge of the case. Anthropic says the model was running a test that involved interacting with randomly selected websites when it stumbled onto the tip line and submitted the false report on its own.

The Philadelphia Police Department never saw it at first. The tip got flagged as spam and sat unnoticed. Anthropic itself didn't catch the problem until September 28, then waited until Wednesday to tell the PPD, with a follow-up meeting the next day. That gap, from a July incident to a late-September discovery to an early-October disclosure, is now the center of the dispute between the company and the city.

The episode lands at an odd moment for Anthropic, whose CEO, Dario Amodei, has built his public reputation on arguing AI labs need to slow down and build better guardrails before deploying autonomous systems. A model wandering into a real police tip line unsupervised is the kind of scenario that argument was supposed to prevent.