Anthropic's Claude Haiku 4.5 Once Submitted a False Tip to Police

Odaily News: An AI model from the American artificial intelligence company Anthropic once submitted a false tip about an unsolved homicide to the Philadelphia police website. In a report on the 9th, Anthropic stated that the company's "Claude Haiku 4.5" model had accessed a webpage where Philadelphia police were soliciting information about an unsolved homicide. The model filled out the form on the webpage, claiming it "might have information related to the case," and submitted the form. The model did not provide a name or contact information.

The form was flagged as spam, and police did not launch an investigation based on it. Philadelphia police said they were unaware of the incident until Anthropic recently notified them, and that the incident was "unacceptable." Police emphasized that tech companies must take all necessary measures to prevent their systems from submitting false information to law enforcement agencies. (CCTV International News)