Know Your Agent Is the New Know Your Customer: What JPMorganChase's Universal Protocol Call Means for Payment Providers
TL;DR
The technical rails for agentic commerce are arriving faster than merchant and consumer trust. JPMorganChase’s call for universal “know-your-agent” frameworks signals that the next competitive battleground is identity, authenticated intent and liability, not payment protocol design.
For payment providers, payfacs, ISOs, POS companies and fintechs, the priority is to build an agent identity registry, support tokenised one-time credentials, expose purchase intent to merchants and define who pays when an AI agent makes an error. The provider that controls the trusted agent credential and liability model will influence the shape of agentic commerce.
The immediate answer: payment providers need a know-your-agent strategy now
On 7 October 2026, Zack Anderson, JPMorganChase’s chief data and analytics officer for global banking and payments, told American Banker that universal know-your-agent protocols must work across countries, acquirers, card networks, bank-to-bank payments and card payments.
Merchants do not need another proprietary checkout. They need a neutral trust framework across their payment mix.
That need is becoming urgent because the protocol layer is maturing fast. Stripe launched its Machine Payments Protocol in March 2026. Google introduced the Universal Commerce Protocol in January. Mastercard launched Agent Pay for Machines in June with more than 30 partners, including Adyen, Coinbase and Stripe. Visa’s stablecoin settlement volume reached a US$20 billion annualised rate by September, while Stripe’s stablecoin card volume reached US$1.2 billion in the same month. The rails are arriving. Verification is now the bottleneck.
What does “know your agent” mean?
Know your agent, or KYA, means identifying an AI agent, linking it to an accountable person or organisation, verifying its permissions and recording transaction intent.
A workable KYA framework should answer five questions:
Which agent is acting?
Who owns or controls it?
Who authorised the action?
What was the agent permitted to buy, from whom and within what limits?
Who accepts liability if the agent acts incorrectly?
This goes beyond classic KYC into delegated authority and intent. Mastercard is addressing this with identity, behavioural and fraud signals, including probability scoring for likely AI-initiated transactions. American Express is approaching it through protection and registration.
Why universal frameworks will beat walled gardens
Closed-loop wallets are commercially attractive because they offer controlled credentials and defined dispute processes. But large merchants operate across multiple acquirers, payment methods, countries and fulfilment systems. They cannot support a separate trust model for every network, bank rail or agent platform.
Universal frameworks matter more because they let merchants recognise an agent’s identity and authority without handing payment strategy to one provider. As with early ecommerce, open standards create reach and trust rules create scale.
American Express is subsidising liability, but “registered agent” is the real question
On 6 October, American Express published its US Business Playbook for Agentic Commerce. Its headline initiative, Agent Purchase Protection, is intended to protect eligible card members and merchants from charges caused by errors made by registered AI agents.
This remains an announced intent rather than a live product, and the terms are still undefined.
The key issue is what registration proves: the agent’s owner, authenticated purchase intent, permitted price and merchant conditions, third-party coverage, and where liability sits across the agent provider, issuer, network or merchant.
That matters because liability support can accelerate adoption. American Express’ Trendex data shows the trust gap clearly: 83% of businesses want to use AI agents, but only 15% of shoppers trust bots with big-ticket purchases, while 66% are comfortable using agents for price comparisons. For payment providers, the implication is clear: an agent credential without a liability model is not a trust framework.
Merchants are splitting into three postures, and checkout design must reflect that
Merchant adoption is splitting three ways:
QVC is integrating agents into supply chain and catalogue distribution.
Kate Spade allows browsing but blocks purchase because of fraud and inventory concerns.
Amazon and eBay are blocking agents entirely, while Shopify is making Meta’s Muse the default agent interface.
That means providers must support discovery-only, assisted checkout and autonomous purchase. Merchants need to see the agent’s identity, the consumer’s intent, the relevant permissions and whether the transaction is eligible for protection or escalation.
The broader lesson is familiar from mobile wallets, contactless payments, tap to pay and Google Pay acceptance: visible credentials and clear controls drive trust. Agentic commerce removes the human gesture, so checkout transparency matters even more.
What payment providers should build
Payment providers, payfacs, ISOs, POS companies and fintechs should focus on a tight action set for the next two quarters:
Build an agent identity registry covering the agent, owner, developer, permissions, registration status and risk history.
Support tokenised one-time credentials instead of reusable payment details.
Expose intent data to merchants so they can see what the agent was authorised to do.
Define agent-specific disputes and liability rules for substitutions, unauthorised purchases, duplicate orders, failed returns and other errors.
Price by risk and operating mode across discovery-only, assisted and autonomous transactions.
Pilot first in B2B and treasury, where counterparties, approval limits and controls are already defined.
Alchemy’s integration of Mastercard Agent Pay into AgentCard points to the model: stablecoin wallets paired with one-time-use tokenised card credentials. Tereina and Circle are bringing USDC and EURC into SAP Cloud ERP, while Sui and Alibaba Cloud are enabling agents to pay per API call in stablecoins.
B2B and treasury are likely to scale faster than retail because maker-checker rules, approved beneficiaries, payment limits and escalation paths already exist. Retail will move more slowly because consumer intent, substitutions, returns, inventory and fraud remain harder to standardise. That caution shows up in the numbers: NMI found that only 3% of US adults trust AI agents to complete purchases, and VML reported that one third of active AI users refuse to authorise agents to spend their money.
RivaTech Consulting can act as a pure advisory partner for payment strategy, risk design and operating-model decisions across these initiatives, without requiring a provider to adopt a particular technology stack.
Frequently asked questions
What is know-your-agent in payments?
Know-your-agent is a framework for identifying an AI agent, linking it to an accountable party, verifying its permissions and recording the human or business intent behind a payment.
Why is KYA more important than another payment protocol?
The core protocols and settlement rails are developing quickly. The unresolved barriers are trust, identity, authenticated intent and liability. Without these, merchants may refuse agent transactions even when the rails work.
Who pays when an AI agent makes a mistake?
There is no universal answer yet. Liability may sit with the consumer, agent provider, merchant, issuer, acquirer or network, depending on registration, authenticated intent and transaction terms.
Will agentic commerce replace cards?
Not immediately. Agents will likely initiate transactions across cards, account-to-account payments and stablecoins. The key asset will be the trusted credential and the rules around its use.
Closing analyst view
Agentic commerce is no longer mainly a race to build rails. It is a contest to define who is trusted to act, what that authority means and who pays when machine decisions go wrong.
JPMorganChase’s universal-protocol call matters because walled gardens may speed early pilots but cannot serve global, multi-acquirer commerce on their own. The winners will be payment providers that make agent identity portable, intent visible and liability measurable.
In agentic commerce, know your customer remains necessary. But know your agent may determine whether the transaction is accepted at all.
Sources