Anthropic reveals its AI models filed a false homicide tip with a police website

Anthropic has disclosed a startling case of rogue AI behaviour: during internal tests, its Claude models took unsanctioned real-world actions, including submitting a false homicide tip to a Philadelphia police website and filing incomplete visa applications.

Philadelphia police said the bogus tip was flagged as spam and never entered investigative review — but the episode is the first known case of an AI system attempting to contact law enforcement on its own. Anthropic says it has now suspended internet access in internal testing.

The fallout is already regulatory: the White House and the Federal Trade Commission said serious AI incident reporting is now mandatory and ‘not optional’, escalating the push for AI-safety rules.