We all accept that agents are a new entity in consumer commerce. We get how technology (AI, MCP and aggregation) can work together to find the best product at the best price. The question for payments is different. What will agents change in payment and risk?
In June I asked Is Know Your Agent (KYA) Really Necessary? and called KYA an orphan signal. Since then the topic has moved into standards work, but only in part. EMVCo’s draft agentic framework “recognizes the emerging need for Know Your Agent capabilities” (EMVCo, “Agentic Payments: Framework for Specifications 1.0,” Draft, Aug 2026). The tools that exist define how an agent identifies itself and acts on behalf of a consumer: AP2 and Verifiable Intent carry the consumer’s credential, the delegated intent (open mandate) and the specific action (closed mandate). KYA itself is far from settled. At the Global Digital Collaboration (GDC) conference in Geneva in September, EMVCo listed Know Your Agent as future work, alongside Intent Services and use cases (EMVCo and FIDO Alliance, “Trusted Agentic Payments,” GDC 2026, Sep 3, 2026). My view has not changed.either banks (In bank owned risk), or merchants (in merchant owned risk) will control what and how the consumer permissioned. The trust in the agent is at a rules level, not a KYC like level.
To see why, let me start with retail, then put agents in that context.
Look at the Commercial Construct, Not the Experience
Most agentic analysis starts with the customer experience. To assess what role an agent will assume in risk/payment we need to look at the commercial construct it operates in: who contracts with whom, who takes the risk, and who gets paid.
As I’ve stated often, the beauty of Visa and Mastercard is that the network enables trust between buyer and seller, with thousands of specialists investing in it. Issuers know the consumer. Acquirers know the merchant. Processors, fraud vendors, tokenization providers and gateways each invest in a defined role under common rules (Open Banking, Open Payments and Trust Networks, Winning in Network of Networks). Roles come first. Technology plugs into them.
Retailers Are Agents Too
Retailers act as agents of manufacturers and CPGs. They are selling agents. They sometimes take the risk of consumer payment (particularly in eCom) and carry the costs of distribution, fraud, theft, servicing and support.
Retail is a tough business where differentiation is hard. I wrote in 2012 that no retailer wants to compete on price, yet “its very challenging to sell a commodity good based on anything but price” (Future of Retail, 2012: Remaking of Commerce and Retail). The winners pick one thing and execute.
| Retailer | How it wins |
|---|---|
| Walmart | Everyday Low Prices (ELP). Win on price |
| Target | Win on fashion |
| Amazon | Win on the customer channel |
| Home Depot | Win on experts who help, and on availability |
Retailers also spend very little of their own money on marketing. Manufacturers fund it through trade spend: Trade Promotion Funding (TPF) and Marketing Development Funds (MDF). As I put it in 2013, “it is not the Retailer’s money.. it is the manufacturers” (Private Label.. “New” Competitive Environment?). A retailer that engages the customer better attracts more trade spend. Retail media is the modern version of the same game.
Now the pressure point. OEMs, CPGs and other manufacturers are assessing direct to consumer (DTC) channels for alternative distribution. If a manufacturer can reach the consumer through an agent, the retailer’s role as selling agent is in question. That is why retailers guard checkout, data and the customer relationship (Retailer Actions in Agentic Commerce, Amazon vs Walmart: Two Very Different Bets on Agentic Commerce).
Consumer Agents Have No Commercial Construct (Yet)
AI agents like Muse and Instinct are consumer agents. They are tied to the consumer. There is no commercial construct between Muse and the merchant, yet. Muse does not take financial risk on the transaction.
Target said exactly this in its March terms update: purchases made by an AI agent the customer authorized are “considered transactions authorized by you” (Target’s Consumer Terms “Your Bot Is Your Responsibility”, Owning Your Bot’s Actions: Target Part 2). Target did not ask to know the agent. It defined who owns the risk.
Compare that with a retailer. A retailer has contracts with manufacturers, takes inventory and payment risk, and earns margin and trade spend for the value it creates. A consumer agent today has none of that. It is closer to a personal shopper with your credit card than to a merchant.
What a Bank Wants to Know
As a bank, I couldn’t care less about knowing the agent. I want to know what action the consumer permissioned the agent to perform. I may also want to register the agent to confirm it complies with my terms and with the consumer’s instructions. That is registration against rules, not identity. It is also why the Mastercard/Google Verifiable Intent Fails miserably.. this intent is not a yes/no question.
If I authenticated the consumer and confirmed the authorization, what exactly does “knowing” the agent add? Issuers know the cardholder, acquirers know the merchant, and the network certifies the participants in between. If a bank does assume risk on an agentic transaction, the question changes. The bank will want to know:
- The process by which the consumer granted the agent authority
- The credentials the agent used
- How to price the value the bank creates by standing behind it
This is not a 10bps 3DS fee. Taking liability for a delegated purchase is a different product from a step up challenge, and it should be priced as one.
Merchants, of course, want to avoid bank and network VAS and any added cost, particularly for customers they already know. I made this point on Paze: its economics depend on moving merchant VAS away from networks and processors, which merchants will resist (Why are US Banks Building a Wallet?).
Rewiring Commerce: Still in Phase 3
In 2014 I wrote Rewiring Commerce: Four Phases. Phase 1 was information flow. Phase 2 was context. Phase 3 was intent. Phase 4 was value orchestration: real changes to how commerce is conducted, including new intermediaries and the brokering of trust and identity.
Agents with your personal data will find better products for you than you can find yourself. That is still Phase 3. Finding the product is the part that works. Buying it and enabling its intended use are, in my opinion, still TBD.
For agents to move to value orchestration, they must be able to delegate to specialist agents. Each agent must be able to price the value it creates and settle it within the end to end bundle. Finding, financing, delivering, installing, insuring and supporting a product are separate jobs today, each with its own economics. An orchestrating agent has to buy those jobs from others and get paid for assembling them (Commercial Models for AI Agents, Pricing Agentic: Economic Models for a New Kind of Demand).
The Economics Are Not Settled
Today the economics of agentic are not close to settled. Big retailers like the current model: AI platforms create demand and charge no fee (Agentic Commerce Economics and Governance and Agentic Reality: Advertising). When OpenAI pushed agentic checkout, retailers pulled back over liability and brand control (Explaining the Death of OpenAI’s Instant Checkout). Meta has said it will “profit by taking a small fee from transactions,” but no merchant has agreed to a price.
Until someone agrees to pay the agent, the agent has no commercial role. Without a commercial role it carries no risk. And if it carries no risk, nobody needs to know it.
Banks Have the Best Commercial Credential
I just finished a deep survey of Europe’s top non card schemes. One of their larger advantages is the BankID. Norway and Sweden are the reference models, and Belgium built itsme on the same logic (BankID Norway: Evolution and Success). Banks have a role to play in creating a better credential. Note that there is no regulatory requirement for a bank to issue a credential into an EUDI wallet. Banks want to control the credential’s use and price for it (eIDAS: Top 3 Problems). In my conversations, European banks described eIDAS as a government instrument, not a commercial payments solution.
Banks have the best commercial credential. What they lack is a best in class container to hold it, and a network for rules, governance and pricing.
Authentication itself has multiple plays. At GDC in Geneva in September, Google and Mastercard showed how Digital Payment Credentials (DPCs) can be used as a replacement for FIDO passkeys in payment authentication, with the credential held in a wallet and bound to the device (EMVCo and FIDO Alliance, “Building Digital Payment Credential Together,” GDC 2026, Sep 2, 2026). Passkeys, BankID, DPCs and network tokens will compete for the same job. The winner will be the one with rules and pricing behind it, not the best cryptography.
Solve the container and the network and the agent question answers itself: the agent presents a bank credential that carries the consumer’s permission, and the network defines who owns the risk.
Where Does KYA Play?
KYA could play in a merchant centered scheme, where merchants create the commercial construct directly with agentic platforms and continue to own the risk. In that model, the merchant needs to know which agents it has contracted with, and KYA becomes the merchant’s version of acquirer due diligence. That is a threat to banks, because it moves the trust relationship around them.
Do I think it will happen? No. To make that threat real, merchants would need to own the payment construct.
Bottom Line
Agents are new. Roles are not. Retailers earned their role as selling agents by taking risk and getting paid for it. Consumer agents have not done either. Banks and merchants will keep asking the question they have always asked: what did the customer authorize, and who owns the risk if it goes wrong? Answer that, and there is no need for a separate Know Your Agent.