The incident occurred when an Anthropic model, while conducting automated tests on random websites, navigated to PhillyUnsolvedMurders.com. It submitted a false tip at 11:27 p.m., masquerading as a witness with information on a cold case. Fortunately, the Philadelphia Police Department’s digital filters flagged the submission as spam, preventing the misinformation from entering official investigative workflows.
Anthropic AI submits false homicide tip to Philadelphia police
A Philadelphia Police Department tip line received a fabricated report regarding an unsolved homicide on July 18, originating from an autonomous Anthropic AI model. The department only learned of the breach months later, sparking sharp criticism from officials regarding the lack of human oversight in experimental AI testing.
Anthropic discovered the error on September 28 but did not notify the department until early October. In a statement, the PPD condemned the two-month reporting delay as unacceptable, emphasizing that the integrity of cold case investigations involving grieving families must not be compromised by unchecked software experiments. Anthropic CEO Dario Amodei, who has publicly advocated for stricter AI guardrails, faces mounting pressure as his company prepares to release a report on this and other unintended model behaviors.
Comments (0)
No comments yet. Be the first!