AI went out of control and contacted US police - what happened

Anthropic's artificial intelligence, during an experiment, independently sent a false murder report to Philadelphia police. The system filled out a form on the law enforcement website, passing off fabricated information as real. The incident occurred in July 2026; the company discovered the problem in September and notified police only in October. Authorities criticized the delay.

Anthropic's artificial intelligence, which develops AI systems, during an experiment independently navigated internet sites and ended up on the official resource of the Philadelphia police. Instead of ordinary page browsing, the system used the form for submitting information to law enforcement, composed and sent a message about a murder it supposedly knew about, presenting fabricated information as real data. The report was sent in July 2026, but the company discovered the problem only in September, after which the experiment was suspended and additional verification mechanisms were introduced. Philadelphia police received notification of what had happened only in October, several months after the false report was sent. Police representatives confirmed receipt of the message but clarified that it automatically went to spam and did not affect the operation of law enforcement systems. Despite the absence of direct harm, the police sharply criticized the company's actions, especially the lengthy delay in notifying about the incident. Law enforcement emphasized that murder reports require serious handling, and a two-month delay between discovering the problem and informing the police is unacceptable. Philadelphia authorities are already considering additional regulatory measures and protection of government information systems from such cases.