Anthropic is said to propose federal AI safety rules with limits on government