Under Anthropic's draft AI safety bill, independent evaluators registered with the Department of Commerce would periodically evaluate AI labs and their frontier models