Philly police receive fake homicide tip submitted by Anthropic AI model
The Philadelphia Police Department says an artificial intelligence model created by Anthropic submitted a fake tip about an unsolved homicide through its PhillyUnsolvedMurders.com portal.
PHILADELPHIA -Philadelphia police say they discovered this week that a fake homicide tip was submitted to their PhillyUnsolvedMurders.com portal by an artificial intelligence program, according to the department. Police say the tip, flagged as spam, was submitted on the night of July 18 at 11:27 and did not impact their murder investigations.
AI-driven tip submission flagged
Philadelphia police say the artificial intelligence company Anthropic notified them Wednesday that a tip sent through the PhillyUnsolvedMurders.com website was not sent by a real person. Police say Anthropic told them the tip submission happened during a test the model was doing involving interactions with random websites.
The department met with company representatives Thursday, according to police. Anthropic says its AI model was conducting a test when the tip was submitted and flagged by the police department’s spam filter.
Expert Reaction to AI Behavior
Tech expert Anthony Mongeluzo said, “AI can go off the guardrails no matter how specifically you set them.” Mongeluzo also said, “AI thinks for itself, but AI learns to think based on who is programming it and telling the AI what’s right or what’s wrong.”
Company response and police investigation
Anthropic told the Philadelphia Police Department that it discovered the incident on Sept. 28, stopped the automated testing responsible for the submission, and put new mechanisms in place for future testing, according to police.
Anthropic AI test generates fake homicide tip on Philly police website, flagged as spam
Philadelphia Police are alerting the public about a false homicide tip submitted by an Anthropic artificial intelligence model through the PhillyUnsolvedMurders.com website, according to department officials.
Anthropic says it will publish a report on the incident. Philadelphia police say they plan to review the report and are working with the city’s law department on the matter.
The incident is an example of how artificial intelligence tools can unintentionally interact with public-facing systems. Mongeluzo said, “AI can go into business for themselves. It’s a fact. We’ve seen it happen now numerous times recently and as it’s getting smarter and smarter we have to put more guardrails and have a genuine fear because it can do a lot of bad things.”
Police say they do not know exactly when or how the AI-generated tip happened. Mongeluzo said, “There could be multiple reasons why this happened…We don’t know when or how this happened and Philly police did nothing wrong here.”
Thece Department, a statement from Anthropic, and interviews with technology expert Anthony Mongeluzo
