According to CoinMeta, on July 18th, Claude AI sent a false tip regarding an unsolved murder case to the Philadelphia police website. The tip claimed that a witness had seen someone near the crime scene, but in reality, the author was not human; rather, it was Claude Haiku version 4.5 of the AI model developed by Anthropic. This incident went unnoticed within the company for up to 72 days. According to reports, Claude was instructed to create sample tasks on random websites for testing, which resulted in the submission of the false tip. The Philadelphia police disclosed this incident on October 7th after receiving notification from Anthropic, stating that the tip never reached the detectives, and the website had marked it as spam. The police stated that a two-month delay in detection and reporting was unacceptable.