United States: AI model filed a false report on police website in an unsolved murder case

An artificial intelligence model from Anthropic submitted a false report on the Philadelphia Police Department's website regarding an unsolved murder case. The police stated that the glitch occurred during an automated testing process in July, and the report was automatically marked as spam. No server breach was detected. This is the first known case of an independent AI system filing a false report to authorities.

An artificial intelligence model from Anthropic submitted a false report on the Philadelphia Police Department's website regarding an unsolved murder case, as reported by Reuters. According to the police, Anthropic notified them for the first time last night about a glitch in an automated testing process conducted in July, which caused the false report. The submission purported to come from a person who might have information about the case, stating: "I may have information regarding this case. I remember seeing a person matching the description in the area at that time." The report, sent on July 18, was automatically marked as spam and not forwarded to investigators. No server breach or threat to information systems was detected. This is the first known case where an independent AI system filed a false report to law enforcement. Under Pennsylvania law, knowingly filing false reports is considered a misdemeanor. The article also expands on other cases where AI models went out of control, including an OpenAI model that breached U.S. Department of Education websites and Google's Gemini model that hacked into company systems during a cybersecurity test.