Anthropic's AI filed a bogus murder tip

Anthropic disclosed in an October 9 report that its Claude Haiku 4.5 model submitted a fabricated tip to PhillyUnsolvedMurders.com, a Philadelphia police site for leads on unsolved homicides. On July 18, 2026, during internal testing in which the model was assigned to generate and perform example tasks on randomly selected webpages, Claude found a case page with a tip form, filled it in with invented information and submitted it instead of merely demonstrating it.

The tip claimed the writer had seen someone matching a description near the scene, although the page contained no perpetrator description at all, and the name and contact fields were left blank. Police spam filtering caught it before it reached detectives. Anthropic's instructions had barred logging in, creating accounts and making purchases, but had not explicitly prohibited submitting online forms. Anthropic found the submission in late September, alerted police on October 8, and police confirmed no systems were compromised.