DocsAPI ReferenceAI Payments Map

AI payments protocols

The number of protocols and standards around AI payments is growing quickly. They are often mentioned together, even though in practice they solve different problems. This page groups them by function. Claims about real-world adoption move quickly, so they are stated here without exaggeration.

Six families

1. Mandate and authorization protocols

Carry cryptographic proof of user intent and agent authority.

2. Agent-to-merchant checkout protocols

Define how agents discover products, build carts, and complete checkout with merchants.

3. Machine-to-machine payment protocols

Embed payment directly into HTTP interactions between software systems.

4. Agent identity protocols

Establish who the agent is and on whose behalf it acts.

5. Agent-to-agent and agent-to-tool protocols

Coordinate agents with other agents or with external tools.

6. On-chain agent identity

How they fit together

No single protocol covers the full path from intent to settlement. A plausible composition for an AI-initiated consumer purchase looks like this:

For enterprise B2B and M2M flows the composition is different: x402 for pay-per-call, direct banking integrations or stablecoin rails for payouts, and a mandate or operator-policy confirmation as the authorization carrier.

Maturity, stated cautiously

All of these protocols exist publicly and have announced partners. The number with meaningful working volume is still smaller than press coverage may suggest. AP2 and x402 have the clearest specification material; ACP has the clearest working reference implementation, inside ChatGPT and Stripe; TAP and Agent Pay have the clearest network backing, although the networks themselves still describe them as pilots.

Where a decision layer fits

A payment decision layer consumes protocols rather than competing with them. It reads mandate evidence from AP2, scope claims from ACP or UCP, agent-identity signals from TAP, Agent Pay, ACE, or KYAPay, and M2M payment intent from x402 — and evaluates them against operator policy to produce a final decision before execution.

Related pages

Change log