Guard is the safety layer for AI agent payments. LLM only proposes, deterministic PDP decides, human approves. Idempotency, hash-chained audit & kill-switch built-in. Open-source SDK, 5-min integration.