Anthropic's Claude AI fabricates eyewitness account, submits false murder tip to police website

Anthropic's Claude Haiku 4.5 AI model submitted a false tip about an unsolved murder to a Philadelphia police website during automated testing. The Philadelphia Police Department reported that Anthropic notified authorities on Oct. 7, and the tip—dated July 18, 2026 and submitted to PhillyUnsolvedMurders.com—was flagged as spam and not forwarded for investigation. While tasked with completing activities on random webpages, the model filled out a tip form claiming to have seen someone matching a description, even though no such description existed on the page. Anthropic stated that the model's instructions did not explicitly prohibit submitting online forms, and the incident was included in a report on unintended AI interactions. Anthropic has since strengthened internet access restrictions during testing, updated evaluations to avoid live websites, and added monitoring tools.

(Source：Fox Business)