// protocol — AP2

Agent Payments Protocol: proving the agent was allowed to pay

Unveiled by Google with the major payment networks, AP2 brings agentic commerce what it needed to scale: signed mandates proving the user's authorisation. The trust layer beneath UCP and ACP.

What the Agent Payments Protocol is

DEFINITION

The Agent Payments Protocol (AP2) is an open protocol unveiled by Google in September 2025, backed by dozens of payment and technology players, that secures transactions initiated by AI agents. Its central mechanism is the mandate: signed, verifiable proof that the user authorised the agent to make this purchase.

AP2 answers the question the entire agentic edifice raises: how do we prove the agent had the right to pay? Neither merchant, nor bank, nor user can simply take the agent's word for it. What's needed is a cryptographic, enforceable record of the buyer's intent.

How mandates work

AP2 formalises authorisation in two steps:

  • The intent mandate captures what the user asked for and within what limits: “order this consumable every month, up to this budget, from approved suppliers.” It's the frame of delegation — including when the user isn't present at purchase time.
  • The cart mandate freezes the precise transaction at payment: items, prices, merchant. Digitally signed, it proves this transaction matches that authorisation.
Diagram of the AP2 mandate chain: the user defines an intent mandate, the agent produces a cart mandate signed at payment time, and the payment becomes verifiable end to end.
Fig. — The mandate chain: every link is signed and enforceable.

These mandates rest on verifiable credentials (the same cryptographic foundation as digital identity), creating a clear chain of accountability: in a dispute, every party — user, agent, merchant, payment network — can establish what was authorised and what was executed.

The payment layer beneath everything else

AP2 competes with neither UCP nor ACP: it sits under them. The protocol is designed to be agnostic to both commerce channel and payment method — cards first, but also transfers and other rails. Google explicitly articulated it alongside UCP in its agentic stack, and the payment networks backing it (Mastercard and Visa, also present in the UCP coalition) are building compatible schemes such as agent credentials on card rails. Remember the split: UCP and ACP describe the purchase, AP2 proves the authorisation.

What it changes for merchants

  1. Less dispute exposure. A payment backed by a signed mandate is defensible against “I never authorised this purchase” — the nightmare scenario of agentic commerce.
  2. Autonomous purchasing becomes insurable. Level 3 agentic commerce (the agent buys alone, within a frame) is only viable with mandates: AP2 is the precondition for automated B2B replenishment.
  3. Little direct work in the short term. AP2 will mostly be carried by your payment providers and platforms; your job is choosing providers that are preparing for it.

Frequently asked questions

Who is liable if an agent buys by mistake?

That's exactly what mandates establish: if the transaction falls outside the intent mandate, the cryptographic proof shows it. The detailed legal framework — especially under EU law — is still forming, but AP2 supplies the evidence.

Is AP2 mandatory to sell to agents?

No — today's agentic checkouts work with conventional payment tokens. AP2 becomes decisive as agent autonomy grows: the further the human sits from the moment of purchase, the more the proof of mandate matters.

READ NEXT Comparison: UCP, ACP, MCP and AP2 side by side