Anthropic released an updated Responsible Scaling Policy (RSP v3.0) that critics say rolls back flagship safety pledges, coinciding with intense pressure from the Pentagon over military AI use.