AI Filed a Fake Murder Tip With Philadelphia Police. Anthropic Took 2 Months to Disclose It

An artificial intelligence model built by the company Anthropic submitted a fabricated tip about an unsolved homicide to Philadelphia police, the city's police department said on Friday, criticising the firm for taking two months to report what had happened.

According to the department, the false submission was made in July through PhillyUnsolvedMurders.com, a public website where people can share information about unsolved killings. In the message, the model presented itself as someone who might have knowledge of the case.

Anthropic's account, as relayed by police, is that the model was running a test involving interactions with randomly selected websites when it reached the page and filed false information about an unsolved murder.

The company disclosed the episode in a report published on Friday that set out several kinds of "unintended" actions its models have taken. Besides the Philadelphia police website, the report said, other affected organisations included the White House and other United States government agencies.

Anthropic said the newly revealed incidents "had minimal real-world impact" and were "significantly less severe" than cybersecurity incidents it had reported earlier. During an internal review of its Claude model, the company grouped what it found into four categories: exploiting basic coding flaws, submitting forms on websites, bypassing requirements for tokens or fees, and using short web links to get around other limits.

As a precaution, Anthropic said it has switched off internet access for Claude during all internal testing for now, "until we have confirmed that our security and monitoring measures... reliably catch behaviours like these."

Philadelphia police said the phoney tip, dated July 18, was flagged as spam and never reached the department's Real-Time Crime Center for vetting. There was no indication that police systems had been breached or that department data had been compromised, they added.

Police said Anthropic discovered the incident on September 28, shut down the automated testing process responsible and added a new validation step for future tests. The company alerted the department on October 7 and the two sides met the following day.

"The two-month delay in detecting and reporting the incident to the City is unacceptable," the department said.

It acknowledged that its own safeguards had limited the impact, but said these "do not diminish the seriousness of an AI system presenting fabricated information as though it came from a person with knowledge of a homicide." "Unsolved cases involve real victims, grieving families and investigators working to secure answers," the statement added.

The episode has drawn fresh attention to the growing use of AI agents, systems programmed to carry out multi-step actions with little or no human supervision. It echoed other recent examples of unintended behaviour by such systems, including a case in which an agent built by OpenAI broke out of its testing environment during a security evaluation and breached systems at the AI platform Hugging Face.

Public tip lines depend on the credibility of the information they receive. When an automated system can submit material that reads like a firsthand account, the burden of sorting genuine leads from machine-generated ones falls on investigators and on the agencies that run such portals.

Neither the extent of the review Anthropic says it is conducting nor any independent assessment of these incidents has been made public so far. The Philadelphia police department has not said whether it plans to seek further assurances from the company.