KYA (Know Your Agent) Framework -- The New Standard for Agent Identity Verification

What Is the KYA (Know Your Agent) Framework?

In March 2026, an e-commerce site detected a wave of suspicious bulk orders. Investigation revealed that an AI agent linked to a legitimate user account had been repeatedly purchasing items the account holder never intended to buy. The account itself was identity-verified. The payment token was valid. But nobody had verified the scope of the agent's authority.

This case illustrates a fundamental reality: traditional KYC (Know Your Customer) alone cannot guarantee the safety of transactions involving AI agents. The KYA (Know Your Agent) framework is a new trust foundation designed to fill this structural gap. Beyond human identity verification (KYC), it provides mechanisms to verify an AI agent's own identity, permission scope, and behavioral patterns.

So why are existing bot management and API authentication insufficient? Traditional bot detection was designed around a binary question: human or bot. In the age of agentic commerce, however, legitimate AI agents execute transactions as human proxies. The challenge is distinguishing malicious bots that should be blocked from authorized agents that should be welcomed -- and for the latter, continuously verifying on whose behalf they act, under what authority, and within what boundaries. This is the essence of KYA.

The Structural Difference Between KYC and KYA

The most important lens for understanding KYA is its contrast with KYC. While it took decades for KYC to become standardized in financial services, KYA demands a fundamentally different architecture.

As this table shows, the critical difference lies in verification timing and continuity. KYC is essentially valid once performed at account opening. KYA, by contrast, runs verification in real time with every agent action. As Sumsub puts it, "Automation itself isn't the problem -- anonymity is." When AI agents execute transactions at scale, the defense depends on continuously binding those actions to a specific human and specific permissions.

The other fundamental difference is liability attribution. Under KYC, liability rests with the individual. Under KYA, a dual structure emerges where both the human delegator and the agent developer bear responsibility. Prove states plainly that "agentic commerce cannot scale without a foundational trust layer that binds every agentic action back to a verified human and a verified authorization event."

Technical Architecture -- The Digital Agent Passport at the Core

The technical heart of the KYA framework is the Digital Agent Passport (DAP), a concept pioneered by Trulioo. The DAP is a lightweight, tamper-proof token that attaches an "identity credential" to every agent-driven transaction.

The DAP lifecycle consists of five steps. Step one verifies the agent developer -- confirming who built the agent and whether the developing entity is real and trustworthy. Step two locks down the agent code, providing cryptographic assurance that the code has not been tampered with after deployment. Step three captures user permission, explicitly recording what the human has delegated to the agent.

These first three steps represent preparation, but KYA's true value emerges in steps four and five. Step four issues the DAP itself, unifying developer information, code signatures, user delegation, and permission scope into a single token. Step five -- continuous lookup -- is what decisively separates KYA from KYC. It monitors agent status in real time, using behavioral telemetry and risk scoring to detect anomalies immediately.

Skyfire's KYA protocol implements this concept using standard JSON Web Tokens (JWTs). Its compatibility with existing OAuth2, HTTP, and JWKS (JSON Web Key Set) infrastructure means it can integrate into existing systems without new infrastructure investment -- a significant practical advantage.

Meanwhile, Mastercard's Verifiable Intent, announced in March 2026, takes a different approach. It consolidates the consumer's identity, specific instructions, and transaction outcome into a single tamper-proof record, generating a cryptographic audit trail. When disputes arise, it can prove whether the consumer truly instructed the agent to execute that specific transaction.

The Competitive Landscape

Multiple major players are entering the KYA standardization race with distinct strategies.

What stands out in this competition is the divergence in positioning. Skyfire is pursuing an open protocol aimed at becoming an industry standard, advancing enterprise integration through its partnership with F5 to embed into application delivery and security infrastructure. Trulioo is penetrating the payment ecosystem core through its joint Digital Agent Passport implementation with Worldpay and its participation in Google's Agent Payments Protocol (AP2).

In contrast, Visa and Mastercard are building agent authentication on top of their existing card network foundations. Visa's Trusted Agent Protocol is integrated with Akamai's edge-based behavioral intelligence and bot protection, and is also aligned with OpenAI's Agentic Commerce Protocol. Mastercard, meanwhile, is prioritizing real-world validation, starting with Australia's first AI agent payment in March 2026.

Sumsub differentiates primarily on "human binding." According to its Identity Fraud Report 2025-2026, multi-step coordinated attacks increased 180% year-over-year, reinforcing its position that continuously binding agent actions to verified humans is the last line of defense against fraud.

No single standard has been established yet, and we are in a transitional period where multiple protocols coexist. The outcome of the Visa-Mastercard standards battle will shape the direction of the entire ecosystem. For e-commerce businesses, the pragmatic approach is to consider flexible architectures that can accommodate multiple protocols rather than over-investing in any single one.

The Regulatory Landscape -- EU AI Act and NIST Standards

Two major developments are driving KYA from voluntary industry practice toward regulatory requirement.

On August 2, 2026, the EU AI Act's high-risk AI system requirements take full effect. The law mandates human oversight for high-risk AI systems, and according to FinanceFeeds, this provision effectively requires human-binding KYA approaches. Mandatory agent disclosure and risk classification will serve as the catalyst for e-commerce businesses to upgrade KYA compliance from optional to essential.

The second force is NIST's "AI Agent Standards Initiative," launched in February 2026. Led by the Center for AI Standards and Innovation (CAISI), the initiative is built on three pillars: supporting industry-led standards development, promoting open-source protocol development, and researching agent authentication and identity infrastructure. In March 2026, NIST published a concept paper on agent identity and authorization, outlining an approach of adapting existing standards like OAuth, OpenID Connect, and SPIFFE for agents.

Australia's AUSTRAC and the U.S. FinCEN are also exploring how to adapt AML/CFT protocols for AI agents, and the regulatory environment is converging rapidly. What e-commerce businesses must recognize is that these regulations affect every company that uses AI agents -- not just developers. Merchants accepting transactions initiated by AI agents may also face requirements for agent authentication and audit trail maintenance.

What E-Commerce Businesses Should Prepare Now

Adapting to the KYA framework is best approached incrementally.

The first priority is making AI agent traffic to your site visible. According to Vouched's findings, AI agents already account for up to 16% of traffic on some e-commerce sites. Without understanding your own reality, you cannot prioritize countermeasures.

Next, bot management policies need revision. As preparation against agent-driven fraud risk, the shift from blanket blocking to selective management -- permitting verified agents to transact while blocking unverified ones -- is now required. Akamai's analysis notes that uniformly blocking all AI bots is no longer a realistic approach.

From there, evaluate whether your existing OAuth2 and tokenization infrastructure can integrate with KYA protocols. Because Skyfire's KYA protocol is designed for JWT/OAuth2 compatibility, many KYA implementations can be deployed as extensions of existing infrastructure.

Finally, preparation from a dispute and chargeback perspective is essential. Cases where "did the consumer actually authorize this purchase?" becomes the central question will inevitably increase for agent-initiated transactions. Leveraging cryptographic audit trails like Mastercard's Verifiable Intent directly improves dispute resolution efficiency.

Conclusion

The KYA framework is foundational infrastructure for the agentic commerce era -- the "trust OS," as it were. Just as KYC became a prerequisite for financial transactions, KYA will become a prerequisite for every transaction involving an AI agent.

The standardization race has only just begun, and no winner has been declared. But as the EU AI Act takes full effect and NIST advances its standards in the second half of 2026, the question has shifted from "whether to respond" to "when and how."