Anthropic pitches AI safety bill that would limit federal model shutdowns

The confidential proposal pairs outside evaluations with narrow catastrophic-risk thresholds and court approval for suspensions. It has not been formally introduced in Congress.