← Articles

Why AI Agents Get Blocked at Checkout, and What Fixes It

Published July 29, 2026Updated July 29, 2026Analysis

Key takeaways

  • An AI agent is usually blocked because it is unidentified, not because it is malicious. Merchant sites cannot tell a legitimate purchasing agent from a scraper, so they block both.
  • The fix has two parts that must arrive together: a verifiable identity, so the merchant knows who sends the request, and proof that a person approved this specific purchase.
  • Open standards exist for the identity part. Web Bot Auth uses HTTP Message Signatures (RFC 9421) and a published key directory, and card network programs such as Visa's Trusted Agent Protocol build on the same idea.
  • Attempts to make an agent look human put it in a contest with detection systems that improve all the time. That approach gets worse as volume grows.
  • Identity, consent, credentials, and execution form an infrastructure program. Agent builders can build it or use a provider, and the same questions apply to both choices.

An AI agent gets blocked at checkout because the merchant cannot identify it, not because it does anything malicious. Merchant defenses cannot tell a legitimate purchasing agent from a scraper, so they block both. The general fix is a verifiable agent identity plus proof that a person approved the specific purchase, so the merchant has something it can check.

This page describes the symptoms, the cause, and the remedy, and it lists the questions to ask of any identity approach.

What does getting blocked look like?

Getting blocked usually looks like a set of unrelated bugs rather than a clear error message. Agent teams often see the symptoms long before they find the cause.

  • The page loads for you and not for your agent: The same URL works in your browser and returns a challenge, an interstitial, or an empty page to the agent.
  • Sign-in stalls on a code: The merchant sends a one-time passcode to a phone or email, and the agent has no legitimate way to continue alone.
  • A challenge appears at add-to-cart or payment: A CAPTCHA or interaction check appears at the steps where merchants concentrate their controls.
  • Prices and options look wrong: A signed-out session shows different prices, no saved addresses, and no real delivery windows, so the agent builds the wrong order.
  • It works, then stops: A flow that passed in testing fails as traffic grows, because volume makes unidentified automation look like an attack.
  • It works on one merchant only: A hand-tuned flow holds at one site, and each new merchant needs the same work again.

All of these symptoms have the same root cause, so a fix for each one alone does not solve the problem.

Why do merchants block a legitimate purchasing agent?

Merchants block a legitimate purchasing agent because, from the merchant's side, nothing in the request separates it from an abusive one. Merchants use bot detection services from companies such as Cloudflare, Akamai, and DataDome, plus one-time passcodes and CAPTCHAs, to defend against credential stuffing, scraping, inventory abuse, and card testing. These controls score traffic, and unidentified automation scores badly.

Question the merchant must answerWhat an unidentified agent request tells the merchant
Who is this client?Nothing verifiable
Is a real person behind this purchase?Nothing verifiable
Did that person approve these items and this amount?Nothing verifiable
Is the operator accountable for this traffic?Nothing verifiable
Is this a purchase or an abuse pattern?Not possible to tell

With this information, blocking is the correct decision for the merchant. A merchant that lets unidentified automation through absorbs real fraud losses. The fix is not to look more human. The fix is to stop being unidentified.

An agent is not blocked for what it does. It is blocked for what it cannot prove.

Does making an agent look more human work?

Making an agent look more human does not work for long, because it puts the agent in a contest with detection systems that are built and improved to find exactly that behavior.

  • It fails under volume: An approach that passes a demo becomes visible at scale, so the flow breaks when the product starts to grow.
  • It never becomes stable: Detection improves all the time, so the work is permanent maintenance.
  • It puts the agent against the merchant: A merchant that identifies the traffic has every reason to shut it down.
  • It does not address consent: Even if the traffic passes, there is no proof that a person authorized the purchase.

The durable approach is the opposite. Make the agent identifiable and make the authorization behind it provable. A merchant can decide to trust an agent that signs its requests with a published key. A merchant can only score anonymous traffic.

What is the fix for agents blocked at checkout?

The fix for agents blocked at checkout is a verifiable identity plus authorized execution. Identity answers who sends the request, and authorization answers for whom and for what. Merchants need both: an identified agent with no proof of consent is still a risk, and a consented purchase from an unidentified client is still unverifiable.

  1. 1

    Sign every request: The agent signs its requests so that a merchant or CDN can verify them independently, with published keys and an open standard such as HTTP Message Signatures (RFC 9421).

  2. 2

    Register where verifiers look: A signature helps only if the verifier can find the key and the operator. Key directories and card network agent programs, such as Visa's Trusted Agent Protocol and Mastercard Agent Pay, turn a signature into a recognized identity.

  3. 3

    Carry proof that a person approved the purchase: The user's request becomes a structured record of merchant, items, and amounts. The user approves it before the purchase, and the system enforces it during execution.

  4. 4

    Keep the card number away from the agent: The user's card is tokenized in a PCI DSS compliant vault, so the agent never sees the number and the builder does not take on card data.

  5. 5

    Execute within the approved terms: The checkout runs against the merchant only within the approved record, so the agent cannot buy outside what the user confirmed.

Each step is possible to build. Together they are an infrastructure program: key management, network registrations, PCI scope, and a consent model that holds when something goes wrong. Builders can do this work themselves or use a provider that does it for many agents.

How does Web Bot Auth make an agent verifiable?

Web Bot Auth makes an agent verifiable by having it sign each HTTP request with a private key and publish the matching public key, so any site can check the signature. Cloudflare proposed the approach, and an IETF working group now develops it.

  • Signed requests: The agent signs parts of each request with HTTP Message Signatures (RFC 9421), for example with an Ed25519 key.
  • A published key directory: The operator publishes its public keys at a known location, so a verifier can fetch them without a private arrangement.
  • Independent verification: A merchant or CDN checks the signature itself, instead of guessing from traffic patterns.
  • Key rotation: The operator replaces keys on a schedule and updates the directory, so a leaked key has a limited life.

Verification is not the same as permission. A verified agent tells the merchant who is asking. The merchant still decides what that agent may do.

Cloudflare, Akamai, and DataDome appear on this page only as examples of bot detection that merchants commonly use. Signed identity does not exempt any traffic from those systems. It gives them a verifiable fact to use in their decision.

What should you check before you trust an agent identity approach?

Before you trust an agent identity approach, check that verification is independent, that the identity is registered where verifiers look, that it covers consent as well as identity, and that card data stays away from the agent.

  • Is the verification independent? A merchant or CDN should be able to confirm the identity alone, with open standards and published keys, without trust in the provider's word.
  • Is the identity registered with the networks? Standing in card network agent programs makes an identity portable across merchants rather than a set of separate agreements.
  • Does it cover authorization as well as identity? Identity alone does not tell a merchant that a person approved this purchase. Look for a recorded approval of specific terms.
  • Where does card data live? If card numbers reach your agent, you take on PCI scope. A tokenized card in a compliant vault keeps the number out of your systems.
  • Who maintains merchant coverage? Ask whether new merchants are your ongoing work or the provider's, so expansion does not become a repeated project.

What changes when an agent has a verified identity and recorded consent?

When an agent has a verified identity and recorded consent, merchant coverage stops depending on how many flows a team has tuned by hand. Identity and authorization travel with the agent instead of being rebuilt for each merchant.

Without identity and consentWith identity and consent
The agent is anonymous traffic to every merchantThe agent sends signed requests that a merchant can verify
Each merchant is a separate engineering projectThe same identity and approval record work across merchants
Card handling puts PCI scope on the roadmapThe card is tokenized and the agent never sees the number
No record of what the user approvedA line-item approval, made before the purchase and enforced during it
Approval sends users to a third-party pageApproval appears in the agent's own interface, often with a passkey

When the charge runs on the user's own card, the user keeps the normal dispute rights of that card.

Blocking is a symptom of anonymity, and anonymity can be fixed. Give the agent a verifiable identity, give the merchant proof that a person approved the purchase, and the agent can spend its effort on what to buy.

Frequently asked questions

Why is my AI agent blocked at checkout when it does nothing abusive?

An AI agent is blocked at checkout because merchants cannot verify that it is legitimate. Bot detection, one-time passcodes, and CAPTCHAs score unidentified automated traffic as risk, and a purchasing agent looks the same as a scraper. A verifiable identity plus proof of user approval addresses the cause.

How can an AI agent prove its identity to a merchant?

An AI agent can prove its identity by signing its HTTP requests with HTTP Message Signatures (RFC 9421) and publishing its public keys in a directory, as Web Bot Auth describes. Card network programs such as Visa's Trusted Agent Protocol build on the same standard.

Does a verified identity let an agent bypass bot detection?

A verified identity does not bypass bot detection. It gives bot detection systems and merchants a fact they can check: which operator sent the request. The merchant still decides whether to allow it and what the agent may do.

Why not make the agent look like a human visitor?

Making an agent look human puts it in a contest with detection systems that improve all the time. The approach fails as volume grows, needs permanent maintenance, and still gives no proof that a person approved the purchase.

Do I have to handle card data to fix the blocking problem?

You do not have to handle card data to fix the blocking problem. The user's card can be tokenized in a PCI DSS compliant vault, so the agent never sees the card number and your systems stay out of card data scope.

What does a merchant need to trust an agent purchase?

A merchant needs two things to trust an agent purchase: a verifiable identity for the agent, and proof that a person approved the specific merchant, items, and amount. Either one alone leaves a risk the merchant cannot measure.

Related articles