# Verifying AI Agents > Detect AI agents and verify attributes about the people behind them, including proof of humanness, eligibility claims, and proof of identity. Source: https://help.withpersona.com/articles/CmxigxTOZVhlr4bn7xSX9S/ Section: Getting started > Use Cases > Verifying AI Agents (KYA) _Last updated on November 7, 2026._ ℹ️ This article explains Know Your Agent (KYA), including how to verify the person behind an AI agent, and the Persona products that can help you do so. There's no one-size-fits-all approach to KYA, and best practices continue to evolve. ## Overview Know your agent (KYA) is how you decide whether to allow a request from an AI agent acting in your product. People use agents to shop, book, and manage accounts for them, so each request needs a decision: accept it, block it, or send it for review. This article covers decisions that depend on the person the agent acts for. ## An overview of KYA inputs Every agent request comes down to three inputs: - **The action:** what the agent wants to do, how much damage a mistake causes, whether you can undo it, and how sensitive the data is. - **The agent:** what it is, who runs it, what it is allowed to do, and how it has behaved before. - **The person:** verified facts about the person the agent acts for. How businesses apply these inputs to agent requests depends on their risk tolerance and use case. Generally, the riskier the action, the more inputs required to decide whether to allow the request. For example, actions with wide reach or that are hard to undo, such as moving money or changing account access, call for verified facts about the person. ## When the person matters Verify the person when a rule, a regulation, or a risk requires assurance that they're human, or about who they are. Even legitimate agents can present risks associated with the person. For example: - **Age and eligibility:** an alcohol retailer must sell only to buyers who are 21 or older, and a car rental must confirm the renter meets age and license requirements. - **Fair access to limited goods:** one person sends many agents, or opens many accounts, to claim more than a fair share. - **Account ownership:** an agent acts on an account its user does not own, or moves money the owner did not approve. - **Role and authority:** an agent does something the employee behind it is not permitted to do. Agent signals alone do not answer these questions, because each agent can look legitimate. ## Layers of identity Different requests need different facts. These four layers answer different questions about the person behind an agent. ### Humanness Is a real human behind the agent? [Selfie Verification](https://help.withpersona.com/articles/1l0WTbR5UsfiggNDPWbnUp/) checks that a live person is present. Liveness detection and anti-spoofing safeguards help distinguish genuine users from automated or synthetic inputs. [Relay's Live Human Presence claim](https://help.withpersona.com/articles/ytHNjFfek8K22wv85UPgVS/) returns this result without handing identity data to your systems. ### Eligibility Does the person meet the requirements for the requested action, such as an age threshold or a jurisdiction restriction? An [age assurance](https://help.withpersona.com/articles/5I4utwPPHF4TilUvZTJCsh/) flow can use age estimation, age inference, or age verification. [Government ID Verification](https://help.withpersona.com/articles/5vXD7S7pQCq8Q9Z4RztxLw/) reads and checks an ID document. [Selfie Age Estimation](https://help.withpersona.com/articles/45UrKCZtpwW5Z7DJ0JW0vM/) estimates age from a selfie. The Relay [Age 21+ Verified claim](https://help.withpersona.com/articles/3O6ZjBIJcLDH1bVoDFCWyA/) returns only whether the person meets that age threshold. ### Uniqueness Is the agent backed by one unique individual, rather than the same person using multiple accounts? Uniqueness establishes that the person is distinct from other users without exposing their underlying identity. A [Live Human Presence result](https://help.withpersona.com/articles/ytHNjFfek8K22wv85UPgVS/) alone does not establish uniqueness. ### Identity proofing Who is the person behind the agent? Identity proofing establishes an authentic, known identity for the person. It confirms that identity evidence is genuine, its details are accurate, and the person owns that identity. [IAL2 identity proofing with Persona](https://help.withpersona.com/articles/QC1OAt4zLihlELoVNJ5jmK/) explains these checks. ## Match the check to the request Use the lightest layer that covers the risk. A lighter check adds less friction and collects less personal information. 1. **Agent signals only,** for low-risk requests like browsing. 2. **A single claim,** such as an age threshold, for age-restricted purchases. 3. **A full identity check,** for actions where a fake or stolen identity is the main risk, such as opening an account. 4. **A full identity check plus the person's explicit approval of the action,** for requests that are hardest to undo. Your product collects and records that approval. When a person has already verified with Persona, [Verify with Persona](https://help.withpersona.com/articles/4Yl6a7VEH2ScPfAHDif0Ts/) turns those past verifications into evidence at onboarding, with their consent. ## Verify the person before you link the agent Use these steps when a task an agent requests depends on who the person is. Skip them for tasks that need no check. 1. **Decide what the task requires.** Determine the fact you need to confirm, such as age or license status, and the lowest level of assurance that is enough. 2. **Choose the verifications.** Most identity checks start with [Government ID Verification](https://help.withpersona.com/articles/5vXD7S7pQCq8Q9Z4RztxLw/). Add [Selfie Verification](https://help.withpersona.com/articles/1l0WTbR5UsfiggNDPWbnUp/) when you need to confirm the person presenting the ID is the person who owns it. 3. **Set up an Account for the person.** An Account ties their verifications to one record. See [Creating Accounts](https://help.withpersona.com/articles/5O2wfeWGy3cR7jS0rdcd1R/). 4. **Build the flow.** Use a Workflow to act on the result, for example to approve the person or send them for review. See [Creating Workflows](https://help.withpersona.com/articles/20Zvcq50493eMUdt7aDhRY/). 5. **Send the person through the Inquiry.** They complete the checks themselves, before the agent acts on the task that needs them. 6. **Link the agent only after the result.** When the Inquiry passes, link the agent to the person's Account in your own product and limit what it can do to what you verified. If it fails, keep the agent restricted for that task. ## What to check when it does not work - **The person is not verified.** Open their Inquiry and confirm it reached a completed state. An Inquiry the person has not finished does not give you a result. - **A verification failed.** Review which check failed and why. If a Selfie Verification failed, confirm the ID was also checked. - **The agent can still act.** Persona does not control the agent. Confirm your product restricts the task until the required criteria are met, such as a passing Inquiry. - **The result does not match the Account.** Check that the Inquiry belongs to the same Account you linked the agent to. If you still cannot find the cause, get in touch and include the Inquiry ID. To learn more or request access, [let us know](https://app.withpersona.com/dashboard/contact-us?category=sales). We're happy to help.