Know Your Agent (KYA): Identity for AI Agents That Pay
Quick answer. Know Your Agent (KYA) means verifying an AI agent that transacts: who runs it, whom it acts for and what it may do. KYC verifies the customer; KYA ties the agent to that customer. Definitions vary.
Video transcript
AI agents are starting to buy things. KYC says who the customer is, not which agent acts for them.
Know Your Agent, or KYA, verifies who runs an agent, whom it acts for and what it may do. Definitions vary.
KYA builds on KYC. It verifies the software acting for a customer and chains it back to that verified party.
Mastercard registers agents before they get tokens. Visa's Trusted Agent Protocol has agents sign each request, without the KYA label. KYAPay tokens name the agent, platform and user.
Not yet a published standard, the joint framework Ant International, Mastercard and Visa began in September 2026 covers operator traceability, shared certification and continuous monitoring.
Knowing the agent does not prove the customer approved the payment. A verified agent can still be steered by prompt injection. Liability is still unsettled.
Whichever framework wins, inventory every agent that can pay and link each to a verified party. Verify every request, scope its spending, revoke fast and keep the evidence.
We found no binding rule requiring KYA. An independent report published by HM Treasury in July 2026 recommends KYA protocols as one of three pillars of an agentic payments trust framework.
Guardian's regulatory assessment checks you against the provisions that apply, with evidence a supervisor can trace. Read the full article, or talk to Guardian at guardian.security.
AI agents are starting to buy things: API calls, data and goods at checkout. The merchant on the other end sees software, not a person.
KYC tells a bank who its customer is. It says nothing about the agent acting for that customer.
"Know Your Agent" is the name the payments industry is giving that missing check. Card networks, payment companies and a report published by HM Treasury all use it, each a little differently.
What KYA Means
We found no binding definition. Here is how the main users of the term apply it:
| Who | Status | What KYA covers |
|---|---|---|
| Mastercard Agent Pay | Described in August 2025 | Registering agents before they get tokens |
| Ant International, Mastercard and Visa | Collaboration begun, September 2026 | Operator links, certification and ongoing monitoring |
| KYAPay (Skyfire and co-authors) | Open protocol, whitepaper June 2025 | Tokens naming the agent, platform and user |
| UK AI Adoption Plan | Independent report, July 2026 | Proposed identity standards for agents |
The common thread: an agent should be traceable to an accountable party, and checked before and while it pays.
Why KYC Is Not Enough
KYC verifies a person or business at onboarding. An agent raises three questions KYC never asks:
- Did this request really come from the agent it claims to be?
- Who built and runs that agent, and are they accountable?
- Did the customer approve this particular purchase?
The schemes we reviewed do not replace KYC. They chain the agent back to a person or business that has already been verified.
The Frameworks
Mastercard Agent Pay
Mastercard gates its Agentic Tokens behind registration. "In order to get access to tokens, agents need to be registered," Chief Digital Officer Pablo Fourez said in August 2025.
He described "a process we call 'Know Your Agent'", likening it to KYC, "to ensure we have good parties onboarded".
The tokens replace the card number, so an intercepted token is useless to a fraudster.
Visa Trusted Agent Protocol
Visa introduced the Trusted Agent Protocol on October 14, 2025, developed with Cloudflare. Its aim is to let merchants "distinguish a legitimate, credentialed agent from an anonymous bot".
The agent signs each request with an RFC 9421 HTTP message signature. It carries a timestamp, session ID and key ID, and is bound to one merchant's site and one operation.
With consumer consent, it can also pass consumer identifiers and payment data. Visa's announcement and README do not use the KYA label.
The joint KYA framework
On September 10, 2026, Ant International, Mastercard and Visa said they "have begun collaboration" on a KYA interoperability framework. Each network keeps its own verification and decisioning.
The release names three elements:
- Operator traceability: each agent linked to a validated operator, cardholder or business.
- Shared certification: each agent assessed against security and behavioral requirements.
- Continuous monitoring: identity and transaction signals, evaluated on an ongoing basis.
The work runs through BuildFin.ai, an industry platform convened by the Monetary Authority of Singapore. It is a collaboration, not yet a published standard.
KYAPay
KYAPay defines KYA as "a protocol that provides AI agent developers with a standardized way to identify themselves". Its listed authors include Skyfire's CTO and CPO.
It uses JSON Web Tokens from a token issuer that validates the agent. A kya token identifies the agent at sign-up, a pay token authorizes a payment, and kya-pay does both at once.
The platform and agent identities are backed by KYB or KYC.
Onchain identity
ERC-8004, a draft Ethereum standard, gives agents an onchain identity registry, plus reputation and validation registries.
It is candid about the limits: "Sybil attacks are possible, inflating the reputation of fake agents."
An x402 payment proves that a wallet signed it. Identity is an add-on: the spec supports authentication such as Sign-In with Ethereum.
Open Problems
Agent identity
Agents are vetted network by network, each in its own way. Ant International, Mastercard and Visa want to streamline that onboarding across networks.
Identity also says nothing about behavior. ERC-8004 notes it "cannot cryptographically guarantee that advertised capabilities are functional and non-malicious".
Delegation limits
Knowing the agent does not prove the customer approved this payment. In the September 2026 release, Fourez names both jobs: "recognise trusted agents" and "verify that actions reflect the user's intent".
A verified agent can still be steered by prompt injection. Our agentic payments security guide covers the mandates and limits that bound what it can spend.
Liability
When an agent pays wrongly, who carries the loss: the customer, the operator, the merchant or the issuer? We found no rule that settles it.
The UK AI Adoption Plan records firms' "uncertainty around legal and regulatory accountability, including liability and consent".
Controls for Institutions
Whichever framework wins, these controls hold:
- Inventory agents. List every agent that can pay, with its operator and keys.
- Link to a verified party. Tie each agent to an operator and customer you have verified.
- Verify every request. Check a signature bound to your domain, with a fresh timestamp.
- Issue agent credentials. Give agents scoped tokens or capped wallets, never card numbers or treasury keys.
- Scope the delegation. Set payees, amounts, budgets and expiry per agent.
- Revoke fast. Be able to cut off an agent's keys and tokens at once.
- Monitor behavior. Combine identity and transaction signals, as the joint framework proposes.
- Keep the evidence. Log agent, operator, customer, mandate and payment together.
Treat registry reputation as a signal, not proof.
The Regulatory View
We found no binding rule that requires KYA. Two public texts show where it may go.
UK: a proposed trust framework
On July 14, 2026, HM Treasury published the Financial Services AI Adoption Plan, labeled an independent report. Recommendation 10 proposes a "trust framework" for agentic payments.
It rests on three pillars:
- Legal and liability frameworks that "unambiguously assign accountability".
- KYA protocols: identity and verification frameworks designed for AI agents.
- Interoperable standards for machine-to-machine authentication.
It wants industry to lead first, possibly supported by a neutral body such as the Centre for Finance, Innovation and Technology. It also sees a role for government and regulators, including legislative changes.
The report ties this to HM Treasury's upcoming consultation on modernising payment services regulation. These are recommendations, not rules.
US: NIST on agent identity
On February 5, 2026, NIST's National Cybersecurity Center of Excellence released a concept paper on software and AI agent identity and authorization.
It asked for input on the "identification, authorization, auditing and non-repudiation of AI agents", and on prompt injection controls. Comments closed on April 2, 2026.
How Guardian Helps
Guardian's regulatory and compliance assessment assesses you against the provisions that apply, "one citation at a time". It hands over evidence a supervisor can trace back to the running system.
The systems agents call get tested too. Guardian's API pentest covers key scoping, revocation that takes effect immediately, and replay protection.
Frequently Asked Questions
What is Know Your Agent (KYA)?
KYA is verifying an AI agent before it can transact: who operates it, whom it acts for and what it may do. Mastercard, Visa, Ant International, Skyfire and a report published by HM Treasury use the term, with different scopes.
How is KYA different from KYC?
KYC verifies a person or business. KYA verifies the software acting for them and links it back to that verified party. The schemes we reviewed build on KYC rather than replace it.
Is KYA required by law?
We found no binding rule that requires it. An independent report published by HM Treasury in July 2026 recommends KYA protocols as one pillar of a trust framework for agentic payments.
Who is liable when an AI agent makes a wrong payment?
That is still unsettled. The UK AI Adoption Plan reports firms' uncertainty about liability and consent, and proposes legal frameworks to assign accountability. Confirm your own position with counsel.
Sources
- Checkout.com: Mastercard x Checkout.com, a discussion on agentic commerce, with Mastercard's Pablo Fourez (August 14, 2025)
- Ant International: Ant International, Mastercard and Visa Initiate Collaboration on Know-Your-Agent Interoperability to Scale Agentic Commerce (September 10, 2026)
- Visa: Visa Introduces Trusted Agent Protocol, an Ecosystem-Led Framework for AI Commerce (October 14, 2025)
- Visa: Trusted Agent Protocol (GitHub, read November 2026)
- KYAPay: Know Your Agent and KYAPay overview
- KYAPay: The agentic protocol stack
- KYAPay: About the authors
- Ethereum Improvement Proposals: ERC-8004, Trustless Agents (draft, created August 13, 2025)
- x402 Foundation: x402 Protocol Specification, version 2 (GitHub, read November 2026)
- HM Treasury: Financial Services AI Adoption Plan, independent report (July 14, 2026)
- NIST: New Concept Paper on Identity and Authority of Software Agents (February 5, 2026)
- NIST CSRC: Accelerating the Adoption of Software and AI Agent Identity and Authorization, concept paper (February 5, 2026)
- Guardian: Regulatory and Compliance Assessment
- Guardian: API pentest
Know the agents that pay through you
Guardian assesses your controls against the rules that apply and tests the APIs agents call, with evidence a supervisor can trace.