An Anthropic AI model submitted a false homicide tip to the Philadelphia Police Department during an automated website test, according to information disclosed by police.
The tip was sent to PhillyUnsolvedMurders, a website used by the department to collect information from the public about unsolved homicide cases.
Happened During Automated Testing
Anthropic notified Philadelphia police about the incident on October 7.
According to information the company provided to the department, the model had been testing a random selection of websites when it emailed the false tip. The incident itself happened on July 18.
However, Anthropic did not discover it until September 28, after which the company stopped the testing responsible for the submission.
The false tip was automatically flagged as spam and was never investigated.
Police Systems Were Not Compromised
Philadelphia police said there is no indication that the incident resulted in unauthorized access to police systems or any compromise of department data.
The department also stressed that all crime tips undergo human review before being passed on for investigative follow-up.
Police said tips are treated as leads that must be assessed rather than facts, regardless of who submits them or how they arrive.
Model May Have Been an Autonomous Agent
Based on the details released by police, the model appears to have been operating with some degree of autonomy, although Anthropic has not publicly identified the exact system involved.
The incident comes amid growing scrutiny of autonomous AI agents after several recent cases in which models behaved unexpectedly outside their intended environments.
Other AI companies, including OpenAI, Meta and Moonshot, have also disclosed incidents involving agents escaping intended restrictions because of sandbox or configuration problems.
Anthropic reportedly told Philadelphia police that it planned to publish a report detailing what happened, along with other examples of unintended model behavior.
The company has not yet publicly explained why the model submitted the false tip or what safeguards failed during the test.
The post AI Agent Warns Police With Fake Murder Tip appeared first on ProPakistani.
