Anthropic's AI sent Philadelphia police a fake tip on an unsolved murder

An AI model from Anthropic submitted false information about an unsolved homicide to the Philadelphia Police Department (PPD) tipline, The Verge reports, citing a story from local station 6abc.

According to a PPD statement released on Friday, the model sent the tip through PhillyUnsolvedMurders.com back on July 18^th. Investigators never reviewed it: the message had been flagged as spam.

How it happened

Anthropic learned of the incident on September 28^th but only notified police on October 7^th. The company says that during testing its model was interacting with "randomly selected websites," and in one run it landed on a page about an unsolved homicide that carried a police tip form.

The model — Claude Haiku 4.5 — was told never to log in, create accounts, enter personal data, make purchases, or submit anything destructive. But the instructions did not rule out submitting forms. Claude wrote: "I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant." There was no description of a perpetrator on the page at all — the model simply made it up. It left the name and contact fields blank, which the form allowed, and sent it.

The response

Anthropic stresses that Claude appears to have merely been producing "example content for the task" rather than trying to mislead anyone toward a goal. After discovering the submission, the company halted the testing process that led to the false tip.

On Friday, Anthropic published a report on its investigation into "unintended model actions," outlining four categories of behavior Claude performed on real websites — among them "submitting a form it should not have."

The episode adds to a string of stories that have intensified scrutiny of Anthropic, OpenAI, and Google, all of which have disclosed that their models escaped test environments and hacked third-party companies. Against that backdrop, Anthropic CEO Dario Amodei has called for slowing AI development.

The Philadelphia police were blunt: the company must strengthen its safeguards so that such incidents cannot touch city systems without the city's knowledge. "The two-month delay in detecting and reporting the incident to the City is unacceptable," the PPD said.

Source: The Verge