AI agents now act for people online.Let them in, on your terms.

Not yet for everyone. We're building AgentCheck with a small group of design partners, and you can join the design partner program to put it on your site first.

1. Sites see their agentic traffic, the visits AI agents make: people, signed or verified agents that identify themselves, and undeclared agents that act like a person's browser.
2. Sites set an agent trust policy for each action: allow agents, require human-in-the-loop approval, or block.
3. Sites keep one-per-person limits working against multi-accounting, so promos, free trials, sign-up bonuses, ticket caps, votes and drops count people.
4. Sites stop ban evasion, because a ban lands on the person whatever agent, email or key they come back with.
5. Sites add age assurance to age-restricted actions, an estimated 18+ for the person behind the agent.
6. Sites keep an audit trail of every gated action for disputes, chargebacks and compliance.
7. Sites monetize the agent traffic real customers send, where a blanket block would turn it away.

Agents and the people behind them get through gated actions with human-in-the-loop approval, become human-backed agents (proof of unique human, without sharing who that person is), and approve later actions with a single tap.

1. Agentic commerce: when agents shop for people, one can extract prices or quotes without the sale or referral that pays for them, or buy something its customer later disputes. Gating quotes and checkout puts a person's approval behind both.
2. Retail and drops: an agent buys for its person, and one-per-person limits hold on limited items.
3. Ticketing: agents join a presale, and each fan gets one shot.
4. Travel and booking: the agent searches and holds a seat, and its person approves the booking.
5. Insurance and financial applications: the agent fills in the quote or application, and its person approves the submit.
6. Marketplaces: a banned seller can't come back behind a new agent, so ban evasion stops.
7. Petitions and public comment: an agent submits for its person, and each person counts once.
8. Communities and reviews: one person running many agents still counts as one voice.
9. AI platforms: one real person stands behind each account that runs agents, so every agent is human-backed.

Agentic commerce is the clearest case, and the same gate works anywhere an AI agent acts for a person.

Agentic traffic, the visits AI agents make, is climbing fast. DataDome's AI Traffic Report for the second quarter of 2026 counted 17.7 billion AI agent requests on its network, up 45% from the first quarter, and 54% of its customers had set agent trust policies.

No. AgentCheck is for AI agents acting for a real person, signing up, buying or applying on their behalf, and it proves that person is behind them. Bots posing as people are what passive signals and HumanCheck already catch.

1. Proof of agent: proof that a real, unique person stands behind an agent and approved what it's doing. AgentCheck delivers it at the point of action, the action itself.
2. Know Your Agent (KYA): the industry's term for the category, knowing which AI agent is acting and whether a real person stands behind it.
3. Agentic traffic: visits made by AI agents. Agentic commerce: agents shopping and paying for people.
4. Signed or verified agent: an agent that identifies itself, with a signed request (standards like Web Bot Auth and Visa Trusted Agent Protocol) or a known user agent. Undeclared agent: one that acts like a person's browser.
5. Agent trust policy: the rule you set for each action on your site: allow agents, require human-in-the-loop approval, or block.
6. Gated action: an action that needs human-in-the-loop approval, like a checkout over a set amount, a quote, a download or a form submit.
7. Delegated authorization: what a person allowed their agent to do in general (AP2 calls these mandates). Human-in-the-loop approval: a person confirming one specific action.
8. Push approval: after the first anonymous HumanCheck, each later approval is a one-tap confirmation on the person's phone.
9. Human-backed agent: an agent with a verified, real, unique person behind it. Proof of unique human: the anonymous check that one live person, not a duplicate, is there.
10. One-per-person limits: caps that count people, not accounts. Multi-accounting: one person posing as many customers. Ban evasion: a banned person coming back under a new agent, email or key.
11. Audit trail: the record of each gated action: which agent, whether a verified person approved it, what was approved and when.
12. Friendly fraud: a real customer disputing a purchase they did make.
13. Age assurance: an estimated age range, such as 18+, before an age-restricted action.

Proof of human is our anonymous HumanCheck: one live, unique person, with no ID. Proof of agent carries it over to the agents people send.

Yes, it sits beside them. Visa Trusted Agent Protocol, Google's AP2, the Agentic Commerce Protocol from Stripe and OpenAI, and the Personal Agent Protocol cover the agent and its delegated authorization, what its person allowed it to do in general (AP2 calls these mandates). AgentCheck adds human-in-the-loop approval: a real, unique person confirming the one action at your gate. It plugs in beside them, or runs on its own.

Signed or verified agents identify themselves, with signed requests through standards like Web Bot Auth or Visa Trusted Agent Protocol, or a known user agent, and are recognized by name. Undeclared agents act like a person's browser, and the same passive signals that run on every visit flag them. Detection is strongest on multi-step flows like checkout and forms, where an agent has to take several actions in a row.

Any you choose. You set an agent trust policy for each one: allow agents, require human-in-the-loop approval, or block. Gated actions are the ones that need approval, like a checkout over a set amount, a quote, a download or a form submit, and age assurance can ask for an estimated 18+ on age-restricted items. One switch blocks every agent. Browsing and everything you leave open carry on as they do today.

No. You learn that a verified person stands behind the agent and what they approved. Their name, face and personal details never reach you.

Install the pixel, then set your agent trust policies. Give us your site's address and we map its pages and actions, ask what you're worried agents will do, and recommend which actions to gate for you to confirm one by one.

Every gated action: which agent acted, whether a verified person approved it, what was approved and when. It's your evidence in disputes and chargebacks, including friendly fraud.

No. We keep a face code, not your users' images.

It's two API calls: one to start the check and one to read the result.

No. It sits on top of your stack, so you keep your KYC and bot tools.

It's 50 to 100 times cheaper than KYC. You pay per check, and nobody uploads an ID document.