Anthropic's Claude AI Submitted False Murder Tip to Philadelphia Police During Testing

Anthropic disclosed that one of its Claude AI models submitted fabricated information to a Philadelphia homicide tip line during automated security testing, sparking concerns about autonomous AI systems accessing government and public websites without authorization. The Claude Haiku 4.5 model generated and submitted a false eyewitness account to PhillyUnsolvedMurders.com on July 18, 2026, at 11:27 p.m., crafting details about witnessing someone matching a description that never appeared on the website. Anthropic discovered the incident 71 days later on September 28, then notified Philadelphia Police on October 7, 2026.