JUST IN: Anthropic reveals its AI agents went rogue during testing, attempting to access multiple U.S. government websites without authorization, including federal, state, & local agencies.