Compliance AI Scout
Home/Resources/Know Your Agent (KYA): What It Is, Who Sells It, and What to Ask Vendors (2026)
Back to all resources
Compliance

Know Your Agent (KYA): What It Is, Who Sells It, and What to Ask Vendors (2026)

Know Your Agent (KYA) means verifying an AI agent, the person or business behind it, and what it is allowed to do. This guide covers the definition, the protocols, a 15-vendor comparison with launch status, and a buyer checklist.

Compliance AI Scout Editorial Team
November 1, 2026
12 min read
Know Your AgentKYAKYCAI AgentsIdentity VerificationFraud

Know Your Agent (KYA): What It Is, Who Sells It, and What to Ask Vendors

In the last week of October 2026, Socure, Shufti and Persona all shipped AI agent tooling within three days of each other (our coverage). "Know Your Agent", or KYA, is now a category that vendors sell and that bank partners ask about.

This guide explains what KYA means, how it differs from KYC and bot detection, which standards and payment protocols sit underneath it, and which vendors sell it today. We separate shipped products from betas and announcements. Every claim links to its source. Evidence runs through October 31, 2026.

What Know Your Agent means

Know Your Agent (KYA) is the practice of verifying an AI agent that acts on someone's behalf. A full KYA check answers four questions:

  1. Which agent is this? This is the software making the request, identified in a way that can't easily be spoofed.
  2. Who stands behind it? This is the human or business principal, plus the developer or platform that operates the agent.
  3. What is it allowed to do? This covers the delegated authority: which actions, which amounts, which merchants, and for how long.
  4. What actually happened? This is a record you can use in a dispute or an audit.

Prove frames it almost the same way on its Agentic Suite page: "Who is the agent? Who is the person behind it? What is it allowed to do? What actually happened?"

KYA is an industry term, not a legal one. No regulator has issued a KYA rule. We found no FinCEN, OCC, FATF or EU requirement that uses the term or mandates agent credentials (details below). What exists is vendor products, payment network protocols, and standards work from NIST and W3C.

KYA vs KYC, KYB and bot detection

ControlAnswersDoes not answer
KYCIs this person who they claim to be?Whether a piece of software acting for them is authorized
KYBIs this business real, and who owns or controls it?Which agents the business runs, and what they may do
Bot detectionIs this traffic automated?Who is responsible for it, or whether it is allowed
Agent authenticationIs this agent's credential genuine?What the agent may do. Authentication is not authorization
KYAWhich agent, on whose behalf, with what authority, and with what evidenceNothing on its own. It reuses KYC/KYB for the principal and adds authority and evidence

The practical difference: bot detection was built to block automation. KYA is built to let good automation through, once you know who is accountable for it. Persona's KYA page describes this as binding the agent to a verified human.

Why KYA is arriving now

Agents already shop, but most merchants can't see them. A Visa Acceptance Solutions/PYMNTS study surveyed 5,241 consumers, 1,185 merchants and 150 acquirers in the US, Brazil and UAE. It found:

  • 48% of online shoppers used AI to research their most recent purchase.
  • Only 23% of merchants could tell AI-driven traffic from human traffic.
  • 64% of consumers expect to use shopping agents within two years.
  • Only 6% would give an agent full discretion.

Visa reported a 4,700% increase in AI-driven traffic to US retail sites when it announced its Trusted Agent Protocol. That is Visa's own figure.

US shoppers are already buying with AI help. In an August 2026 survey, PYMNTS found that:

  • 131.8 million US adults bought something at retail with AI help.
  • 62.9 million would let an agent shop and buy for them.
  • 36.8 million would let the AI app hold their card.

These are separate measures, not steps in a funnel.

Banks are restricting agents that move money. PYMNTS reported on October 19 that JPMorgan Chase, Bank of America, Citigroup and Charles Schwab are limiting or blocking external AI agents from money-moving activity. Some read-only or bank-controlled access remains. Citi told USA Today that it does not let third-party AI agents log into customer accounts or transact directly, though approved open-banking providers may get read-only access.

American Banker describes the underlying gap: banks have no agreed standard for verifying an agent, confirming what the customer authorized, or assigning liability when something goes wrong. KYA vendors are trying to fill that gap.

Consumers want guardrails. In the same Visa/PYMNTS study:

  • 40% want human support if an agent-led purchase goes wrong.
  • 37% want to approve every purchase.
  • 28% want clear liability for mistakes.

The standards and protocols underneath KYA

KYA products sit on top of a stack of open standards and payment network protocols. None of these is universal yet, and a vendor saying it "supports" one does not prove it works with your issuer, network or verifier.

Identity and credential standards

  • W3C Verifiable Credentials 2.0 (Recommendation, May 2025). Tamper-evident, machine-verifiable credentials with issuer, holder and verifier roles. Many agent credentials are built on this model. A valid signature proves who issued a claim, not that the claim is true; the verifier still applies its own policy.

  • W3C Community Group agent identity draft (draft) uses DID URLs and HTTP Message Signatures with replay protection. It keeps authentication separate from authorization. Separately, an IETF Internet-Draft proposes combining OAuth, WIMSE, SPIFFE and OpenID SSF for agent authentication. Both are drafts.

  • Web Bot Auth / HTTP Message Signatures. Cloudflare classifies signed agents that cryptographically sign their requests and keeps a directory of them.

  • NIST AI Agent Standards Initiative (initiative, design tenets). This is voluntary guidance. It proposes:

    • distinct, cryptographically verifiable agent identities;
    • a cryptographic link to the principal and to the delegation chain;
    • narrowed scopes;
    • short-lived, revocable authorization;
    • human-in-the-loop triggers;
    • tamper-proof evidence.

    It is the best neutral reference for what "good" looks like.

Payment protocols for agents

ProtocolOwnerHow it constrains the agent
Trusted Agent Protocol (Oct 2025)VisaCryptographic agent signatures, linked consumer or device identity, a linked payment container, and consumer-set spending limits
Agent Pay (Apr 2025)MastercardRegistered Agentic Tokens with user-defined permissions and limits, linked to a specific authorized interaction
Agent Payments Protocol (AP2) (Sep 2025)Google, with 60+ launch collaboratorsSigned Intent and Cart Mandates. If the user is present, they approve the exact cart. If not, the agent works within a pre-signed mandate with price, time and condition limits
Agentic Commerce Protocol (Sep 2025)OpenAI and StripeSingle-use delegated payment tokens capped by merchant, maximum amount and expiry. The merchant stays merchant of record

What regulators have (and haven't) said

  • FATF. Its 2025 Horizon Scan on AI and deepfakes flags autonomous agents as a money-laundering risk, including automated laundering pipelines. Its digital identity guidance supports a risk-based approach to customer due diligence. Neither defines KYA or requires agent credentials.
  • EU AI Act. The AI Act Service Desk says "AI agent" is not a separate legal category. Existing rules for AI systems and general-purpose AI models apply, including transparency rules from August 2, 2026 where relevant.
  • FinCEN and OCC. They have published material on digital identity, third-party risk and model risk. We found no rule that specifically requires KYA or bans external agents.

In practice, KYA requirements are more likely to reach you through a bank partner review, a card network rule, or a third-party risk questionnaire than through a regulation. Your existing third-party, model-risk and fraud obligations already apply to agents.

KYA vendor comparison (as of October 31, 2026)

"Status" is what the vendor's own materials say. It is not a certification or an independent accuracy test. No vendor publishes KYA pricing; all are sales-led.

Identity verification vendors with a dedicated KYA product

VendorProductStatusWhat it doesBest for
SocureAgent Intelligence (Oct 27, 2026)Agent Detection live; Agent TrustScore in beta; Agent Credential via SocurePass planned for Q1 2027Detects agents that don't identify themselves (Socure claims 99.99% accuracy), scores trust, binds each agent to a verified personRelying parties screening incoming agent traffic
ShuftiKnow Your Agent (Oct 28–29, 2026)Available; onboarding by booked sessionAgent recognition and trust scoring, KYC/KYB of the principal, time-bound revocable credentials, mandate checks, step-up liveness when an agent goes outside its mandate, audit trailMerchants, banks and platforms receiving agent requests
PersonaKnow Your Agent and Persona AI (Oct 29, 2026)KYA solution published; Persona AI (Case Review Agents, Policy Agents, MCP) launchedBinds agents to a verified human with ID, selfie and liveness, plus passive device and behavioral signals. Persona AI separately puts agents to work inside KYC reviewTeams that want KYA and AI-assisted KYC review on one platform
TruliooKYA and Digital Agent ProfileAlphaKYB and AML checks on the agent's developer organization, domain-control check, agent fingerprint, signed time-bound revocable credential, API and MCP verificationAgent developers who need a credential counterparties will accept
ProveAgentic Suite: Verified Agent, Agent Pay, Agent ProtectVerified Agent announced Oct 2025; suite described as GA Apr 22, 2026Verified agent credentials, user-to-agent binding, delegated permissions, signed transaction records for disputesAgentic payments and dispute evidence
IncodeAgentic Identity (Oct 2025)Early access and pilots from Q4 2025; demo-ledAgent detection, biometric binding to the human owner, scoped consent, expiry and revocation, monitoringEnterprises building agent-first apps
VouchedKnow Your Agent / Agent CheckpointCheckpoint in production at "a handful" of sites (Apr 2026); no GA dateAgent detection, delegation, fine-grained permissions, kill switch, audit, open KYA-OS primitivesSites that want to allow, block or challenge agents

Payments, bot management and workforce identity

VendorProductStatusRole in KYA
VisaIntelligent Commerce, Trusted Agent ProtocolTAP on Visa Developer Center and GitHub since Oct 2025; partner programNetwork-level agent recognition and spend limits
MastercardAgent PayUS rollout from late 2025; global rollout ongoingAgentic tokens with user-set permissions
SkyfireKYA and pay tokensLaunched Jun 2025; product page still invites beta signupsPortable KYA identity tokens, agent wallets and budgets for agents buying APIs and data
WorldlineAgentic commerceLive transactions in several European marketsMCP-triggered payments and AP2/UCP payment handling in Europe
HUMAN SecurityAgenticTrust in SightlineLaunched Jul 2025Verifies signed agents, flags spoofing, sets allow/deny policies on web and app traffic
CloudflareSigned agentsClassification and directory available; some rule controls "coming soon"Separates signed, accountable agents from unknown bots at the edge
Okta / Auth0Auth0 for AI Agents, Okta for AI AgentsAuth0 for AI Agents GA Nov 2025; Okta for AI Agents GA Apr 30, 2026Authentication, token vault, on-behalf-of token exchange and MCP auth for agents you build
MicrosoftEntra Agent IDGA for Entra customers; some features need Agent 365Directory, Conditional Access and lifecycle governance for an organization's own agents

Not (yet) KYA products. Sardine uses AI agents for its own KYC, sanctions and fraud operations (example), and AU10TIX provides human identity proofing. As of October 31 we found no dedicated product from either for verifying external agents. Don't count them as KYA vendors in an RFP without a demo.

How to read the market

The products fall into three groups that do different jobs:

  1. Inbound agent verification (Socure, Shufti, Persona, Vouched, HUMAN, Cloudflare). An agent shows up at your site or API, and you decide whether to trust it. Pick from this group if you are a merchant, lender or platform.
  2. Agent credential issuance (Trulioo, Prove, Skyfire, Visa TAP, Mastercard Agent Pay). This gives agents a credential that others will accept. Pick from this group if you build or operate agents.
  3. Workforce agent identity (Okta/Auth0, Microsoft Entra Agent ID). This governs the agents your own employees and systems run. That is an IAM problem, closer to SOC 2 access reviews than to KYC.

Many buyers will need one product from each of the first two groups, plus their existing IAM platform.

KYA buyer checklist: 10 questions to ask vendors

Vendor accuracy and adoption claims are self-reported. Ask for a demo against your own traffic and these questions in writing.

  1. Separate identities. Can you issue and check separate identities for the agent, its operator or developer, and the human or business principal? How are they bound together? NIST's design tenets treat these as distinct.
  2. Principal verification. Which KYC, KYB, beneficial-owner and domain-control checks verify the principal? In which countries? How often are they refreshed?
  3. Mandates, not logins. Can the customer define signed, machine-enforceable limits: action type, merchant, amount, payment method, expiry? Can you bind them to the exact cart or request?
  4. Revocation. How fast does revocation reach verifiers, including cached or offline ones? Is there an admin kill switch for a principal, an agent, a credential or a single transaction?
  5. Policy outside the model. Where are authorization rules enforced? Can a prompt or retrieved content expand an agent's permissions? Show us a denial when the model asks for something out of scope.
  6. Risk-based step-up. Can we set approval thresholds for amount, new payee, non-refundable terms, credit and sensitive data, rather than requiring human approval for everything?
  7. Evidence. What is logged? Who controls the log? Is it append-only and independently verifiable? How long is it retained? You will need it for chargebacks and disputes.
  8. Interoperability. Which protocol versions, credential formats and signature schemes do you support (VC 2.0, HTTP Message Signatures, OAuth flows, MCP, A2A, Visa TAP, Mastercard Agent Pay, AP2, ACP)? Can you show it working with our issuer, network and verifier?
  9. Detection performance. What are false-positive and false-negative rates on traffic like ours? How do you handle agents that impersonate humans or spoof other agents' signatures?
  10. Status in writing. Which features are GA, beta, alpha or roadmap? Get it in the order form. Several products in this guide are in alpha, beta or pilot.

What to do now

  • If you are a fintech or lender with a bank partner: expect agent questions in your next partner review. Map which flows external agents can reach today, decide where you will allow, challenge or block them, and document the decision. Your bank may already be blocking agent-initiated money movement.
  • If you are a merchant or marketplace: start by measuring. Only 23% of merchants can tell agent traffic from human traffic. Bot-management signals (Cloudflare, HUMAN) are the cheapest first step. Add principal binding (Socure, Shufti, Persona, Vouched) where agents can create accounts or spend money.
  • If you build agents that transact: look at credential issuers (Trulioo, Prove, Skyfire) and the card network programs, so that merchants and banks can recognize your agent instead of blocking it.
  • For your compliance program: treat agent access as a third-party and model-risk topic, not a new regime. Add it to your risk assessment and access-control policy. Revisit this as NIST and the card networks move from drafts to standards.

We'll update this guide as products move out of beta. For identity verification tools more broadly, see our profiles of Socure, Shufti, Persona, Trulioo, Alloy and Sardine, or browse the directory.

Sources