Tech »  Anthropic drops its signature safety promise and rewrites AI guardrails