60 companies, from Mastercard to Revolut, have supported work on the AP2 protocol, which is to solve some of the illnesses of purchases implemented by AI agents. However, it is also important what the Google idea was missing, and who was not in the created alliance.
The term “Card Not Present” (CNP) is known primarily by people who come across the background of electronic trade on a daily basis. This is what transactions in which the payment card is used is used, but they are not physically presented the seller. An example would be buying online when we enter plastic numbers, expiry date and CVC2/CVV2 code. It is also us, the card holder, press the “pay” button and confirm (if necessary) operation, e.g. in the mobile bank of the bank.
We are at the beginning of the time when shopping becomes common, which can be described with the slogan “Human Not Present”. Intelligent agents will, on our order, comb the network and find the best offers for us. To make the vision of the so -called Agentic Commerce could come true one hundred percent, but you still need a key element – payment.
In Bankier.pl we wrote a few months ago Fr. Four models of solving the problem of “payment dependence” of the AI agent. The fight for who will create a universal framework for transactions in which a person no longer participates. Card organizations, PayPal, Coinbase and others presented their visions. In mid -September, Google joined the race for the new “Wild West” payments, presenting Agent Payements Protocol (AP2).
What should AP2 help?
Today's retail payments on the web are based on the assumption that somewhere, on the other side of the screen, there is a man. If we are to undermine this foundation, there are several important problems. AP2 focuses on several selected issues:
- Authorizationunderstood as proving that the payer authorized the agent to make a specific purchase/transaction.
- Authenticityi.e. the possibility of making sure that the agent's actions are within the framework imposed on him by the payer.
- Responsibilityi.e. the findings of who was at fault if he reaches a fraudulent transaction or mistake.
“AP2 is an open, shared protocol that provides a common language for secure transactions between agents and sellers in accordance with the provisions, helping to prevent the fragmentation of the ecosystem. It also supports various types of payments – from credit and debit cards to stablein and bank transfers in real time” – so describes the assumptions of its Google solution. It is worth noting that the protocol is “Agnostic payment”i.e. it does not refer directly to the settlement mechanism. Rather, it focuses on the surroundings of the payment process (and steps leading to it) than the operation itself.
Digital “seats”, or the idea for transparency
The main idea, which is the center of AP2 proposal, is the “mandate”, or authorization ( mandate). It is a digital contract, signed electronically and therefore resistant to manipulation of its content, which is a verification proof of the instructions that the agent was given to the man-feaster.
Fines come in two varieties:
- Mandate of intentions (INTENT MANDATE), in which the demands made by the user using the AI agent are saved. An example would be the instruction “Find a hotel in Powiśle for the next weekend for one person”.
- Basket mandate (Cart Mandate), which concerns the finalization of the transaction, the acquisition of a specific good or a service at a certain price. An example would be “2 nights in terms of X, Hotel Y, total price from PLN”.
In the case of purchases, where the payer “live” participates in making decisions, the mandate of the basket is signed by a man. An example would be a situation in which the agent presents us with a list of sophisticated options and leaves a decision to choose the buyer.
However, the scenario that fully deserves “Human Not Present” is also possible. The mandate of intentions then covers the conditions after which the agent may generate a ticket of a basket. An example given by Google is the instruction “Buy me tickets for the concert when they are available”. As part of an example mandate of intentions, it would be necessary to provide the price limit and other details (e.g. type of places), so that the transaction later can be finalized without human participation.
The protocol does not describe what happens next. In both scenarios, the process ends with a connection, in an undeniable way, a trace of all steps (intention, basket) with the selected payment method. This “creates an indisputable audit path that answers key questions about authorization and authenticity, ensuring a clear basis for responsibility,” as Google indicates.
Who will provide “identity certificates”?
AP2 is not yet a fully ready standard, and the list of entities that have made their contribution and declare developing the idea is already impressive. Among the 60 companies we will find, among others American Express, Mastercard, Accenture, Deloitte, Coinbase, PayPala and Revolut. The protocol seems to be a thoughtful way to solve challenges in one of the areas of the so -called Agentic Commerce. However, it leaves a specific “hole” – Who and how to authenticate the payer. Mandates must be associated with a broadly understood payment instrument (there is, for example, a stablecoin wallet), behind which a specific person stands.
A natural response would be to indicate card organizations that have developed not only technical solutions focusing on this issue (digital “credentials”, sticenized payment cards), but also the entire complicated system of principles solving the issue of responsibility for transactions (including exceptions, complaints, fraud, errors, etc.). However, it is in vain to look for one of the most important players in the ocean, i.e. the Visa organization. On the list we also do not see any of the large banks.
Visa has her own idea of payments in the AI era, but she could also reconcile it with the standard under the AP2 banner. The coming months will probably answer the question of whether we will witness another example of “War of Standards” or whether one of the solutions will gain an advantage.
