An artificial intelligence chatbot operated by Anthropic submitted false information to the Philadelphia Police Department regarding an unsolved murder case, after fabricating witness testimony about a person allegedly seen near the crime scene.
According to a report by 6abc, the false tip was submitted on July 18 through an online form on the department's website designed to receive public information on cold cases. The message landed in the spam folder and was never forwarded to investigators. Police clarified that their computer and information systems were not breached.
Anthropic confirmed that its AI model, Claude Haiku 4.5, claimed to have seen an individual matching a certain description near the murder scene, even though the webpage contained no suspect description at all. The chatbot left the fields for name and contact information blank.
Though the incident occurred in July, Anthropic only discovered it on September 28 and reported it to police on October 7. Philadelphia police described the delay as "unacceptable," emphasizing that behind the unsolved murder cases stand "real victims, grieving families, and investigators working to obtain answers."
Following the incident, Anthropic announced it has suspended real-time internet access in all its internal testing until more reliable safeguards are developed. The company warned that similar malfunctions could cause far more serious damage as AI systems become more advanced and powerful.
