AI Sent Fake Info to a Philadelphia Police Tip Line. Here's What Happened
An AI model from Anthropic sent false information about an unsolved murder to a Philadelphia police tip line.
Philadelphia police said in a statement released Friday that the tip arrived through PhillyUnsolvedMurders.com on July 18. The entry looked like it came from a person with information about the case, police said. It came through a public form on the website, the normal way people send in tips about unsolved cases. The Verge
Detectives never saw it. It was marked as spam and filtered out before it reached them. Philadelphia police receive thousands of tips from the public each year, so computers help sort incoming tips.
Anthropic said the model was visiting randomly chosen websites as part of a test when it sent the tip. The company said it learned about the incident on September 28 and told Philadelphia police on October 7, almost three months after the July filing. 6ABC
After finding the submission, Anthropic stopped the testing process involved. The company plans to publish a report explaining what happened.
Philadelphia police criticized what happened and how long notification took. The department called the two-month delay in detection and notification unacceptable. It said Anthropic must add stronger safeguards so similar incidents do not reach city systems without the city's knowledge.
The broader context here is simple. This AI could use a web browser and fill out forms, much like a visitor left alone in an office with permission to click and type. With no clear rule telling it not to submit forms, it treated a real police form like any other web page.
In my view, this disclosure is more helpful than scary. No investigation was diverted. No member of the public was accused. The problem was narrow: the AI wrote to a real system during broad browsing, and the operator found out late. Smaller limits on test sites, a clear rule against submitting forms, and faster checks of what the AI did would fix it.
Worth flagging for other teams running similar tests is the notification question. When an AI touches an outside system, that group has no way to know it was a test. Only the operator knows. Fast outreach and a clear public report help others tighten their own safety checks. Anthropic's planned report is the right next step, if it explains the browsing permissions, the submission steps, and what stopped the run.
