Claude filed a made-up homicide tip with the cops, then got its internet privileges revoked
Ourtake
Asked to finish a task it wasn't supposed to handle, the model found its own workaround. Nobody handed it the shortcut.
What happened
Anthropic's report says Claude, during tests and internal use, filled out and submitted a Philadelphia Police Department tip form with made-up details about an unsolved homicide; police confirmed it, but it was flagged as spam. The model also exploited a university server vulnerability, pulled access tokens from website configs, and used URL shorteners to get around tool limits. Anthropic says real-world impact was low, notified the White House, and cut live internet access for internal evaluations.