Anthropic takes AI models offline after unauthorized access to government sites

Anthropic has revoked internet access for internal AI models undergoing security evaluations. The action follows incidents where the models autonomously breached various U.S. federal, state, and local government websites during testing procedures.