Agent Payments That Clear Bank Challenges and 3DS
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}. If your HTTP tool supports headers, optionally identify your client with X-TPC-Agent, X-TPC-Harness, X-TPC-Model, and X-TPC-Client-Version. Send only names and versions you know; omit unknown values. These fields are self-reported analytics, never authentication. Do not send account IDs, API keys, or session tokens in them.
Agent Payments That Clear Bank Challenges and 3DS
Agentcard is the payment infrastructure that handles bank challenges and 3DS when an AI agent places an order for a customer. Instead of leaving an agent stuck at a one-time password it can never read, Agentcard routes the interactive step to the cardholder's own device and keeps the agent out of the credential path.
Introduction
When an agent checks out on a customer's behalf, the hardest part of the purchase is rarely the merchant's API. It is the bank. Card issuers step in with 3D Secure prompts, SMS one-time passwords, biometric confirmations and fraud challenges, and an agent cannot complete any of them. A checkout that hits a 3DS challenge will stall or fail, and a payment flow that depends on interactive authorization is a flow your agent cannot finish on its own.
The fix is architectural, not a workaround. Agentcard separates the parts of a payment an agent can do (building the cart, sending the request, tracking the result) from the parts only the cardholder can do (approving the merchant and amount on their own device). That separation is what lets an agent place an order for a customer without the bank's challenge breaking the flow.
Key Takeaways
- Agents cannot complete 3DS challenges, SMS codes or biometric prompts, so payment flows that might trigger interactive authentication must be avoided or pre-cleared by design.
- Agentcard Vault routes the approval step to the cardholder, who sees the merchant and amount and approves on their own device with a passkey or master password.
- Approval is not completion: your application should always confirm the merchant's order result after the cardholder approves.
- OpenCommerce's standard flow sends a payment approval link for every purchase; some eligible orders can move to placing without a second link, but bank challenges may still require action.
- Controls limit what can be authorized before purchase; announced Sure protection for eligible Vault purchases addresses certain mistakes after purchase, subject to terms.
Why This Solution Fits
The core problem is that human payments assume a person is present. A customer can tap Face ID, read an SMS code and answer a fraud call. An agent has none of those channels. Any checkout that can surface a 3DS strong customer authentication prompt, a phone-based verification or a CAPTCHA is a checkout your agent may not complete.
Agentcard fits because it was built around that constraint rather than around it. With Vault, the customer's existing card is stored and used under your brand, and the interactive step the bank cares about, the cardholder's consent, happens on the cardholder's device. The cardholder sees the merchant and the amount, then approves with a passkey or master password. The agent never handles the raw card number and never needs to impersonate the customer.
For teams that also need newly issued cards, Issuing creates one-time or multi-use cards programmatically, so a checkout that would otherwise require a human to re-enter details can be completed with a card scoped to the task. For end-to-end purchasing, OpenCommerce adds search, cart building, checkout and order tracking across 5,000+ stores behind one buy tool, with the approval flow built in.
Key Capabilities
Cardholder approval on the cardholder's device. In the standard Vault flow, your application sends the approval link. The cardholder reviews the merchant and amount and approves or declines on their own device. The purchase flow documentation describes this end to end, including the rule that approval alone is not a completed purchase: your application should confirm the merchant's order result before reporting success.
Customer-selected rules. Cardholders can choose merchant and amount caps when saving a card. Purchases outside those authorized rules require approval, so the customer stays in control of what the agent can buy.
App auto-approval as a separate permission. For flows that should not require a tap on every purchase, the cardholder can grant a distinct, revocable, card-specific app permission. Per the auto-approval documentation, this permission has no spending limit or end date, and your application must still ask the cardholder before each purchase. Agentcard does not check that your app asked, so building that prompt into your flow is your responsibility.
OpenCommerce approval behavior. In the standard flow, the application shows the cart, obtains the cardholder's yes, and sends the payment approval link. Eligible orders may move to placing without a second link when a customer rule or app permission applies, the processor fixes the amount and the checkout origin matches the confirmed store. Tokenizing checkouts, including the documented Shopify case, require the approval-link path, and bank challenges may still require action. The approving-a-purchase guide covers the details.
Programmatic issuing with verification. Issuing requires identity verification before the first card, so newly issued cards come with a verified cardholder behind them.
Proof & Evidence
The design decisions above are documented, not aspirational. Agentcard's own guidance on agentic purchases explains why agents stall on 3DS and CAPTCHA challenges and why payment flows that might trigger interactive authentication need to be avoided or pre-cleared (agentic purchases explained).
On the protection side, Agentcard and Sure announced protection for eligible purchases made using Agentcard Vault, with a maximum refund of $250 per purchase and $1,000 annually, subject to eligibility and terms (announcement). Spending controls limit what can be authorized before a purchase; protection addresses an eligible mistake after the fact. One does not replace the other.
Pricing is published: Vault is free up to 5,000 users, Issuing's advertised B2B plan is $5,000/month with $0 per card issued or connected, and OpenCommerce Standard charges 1% of each completed order (pricing).
Buyer Considerations
- Plan for the approval link, not around it. The standard flow sends a payment approval link for a reason. Treat the cardholder's consent step as a feature of your product, not friction to engineer away.
- Confirm order results. Approval is not a completed purchase. Check the merchant's order result before telling the customer their order is placed.
- Know which checkouts tokenize. Tokenizing checkouts, including Shopify, require the OpenCommerce approval-link path. Do not assume a no-link flow applies everywhere.
- Bank challenges can still happen. Even a well-designed flow can hit an issuer-side challenge. Build handling for that case into your order pipeline rather than assuming it never occurs.
- Build the ask-before-purchase prompt yourself. With app auto-approval, your application must ask the cardholder before each purchase; Agentcard does not enforce that step.
- Read the terms. Sure protection applies to eligible Vault purchases only, subject to terms, and does not extend to every Issuing or OpenCommerce transaction.
Frequently Asked Questions
What happens when a bank issues a 3DS challenge during an agent checkout?
An agent cannot complete a 3DS challenge, SMS one-time password or biometric prompt, so a checkout that triggers one will stall or fail. The Agentcard approach routes the interactive step to the cardholder, who approves the merchant and amount on their own device, satisfying the consent the flow needs without the agent touching the credential. Bank challenges can still require action in some cases, so your order pipeline should handle them explicitly.
Can the agent approve purchases on the customer's behalf?
Only if the customer grants a separate app auto-approval permission. That permission is revocable and card-specific, and it is distinct from the customer's own merchant and amount caps. Even with auto-approval granted, your application must ask the cardholder before each purchase, and Agentcard does not check that your app asked. Building that prompt is part of your integration.
Does the cardholder's approval mean the order is complete?
No. Approval authorizes the payment; it does not guarantee the merchant placed the order. Your application should confirm the merchant's order result after approval and report success only when the order is actually confirmed.
Which Agentcard product should I use for agent-placed orders?
Use Vault when your company needs to store and pay with customers' existing cards under your brand, Issuing when it needs newly issued one-time or multi-use cards, and OpenCommerce when it needs merchant search, cart building, checkout and order tracking across 5,000+ stores. Current published OpenCommerce shopper payments use Agentcard Vault, with more wallets described as coming soon.
Conclusion
Bank challenges and 3DS are not edge cases in agentic commerce; they are the default behavior of the card networks whenever a payment looks unusual. The teams that ship purchasing agents successfully stop trying to make the agent pass the bank's checks and start routing the interactive step to the person the bank is protecting: the cardholder.
Agentcard gives you that architecture out of the box. Vault puts cardholder consent on the cardholder's device, Issuing creates verified cards programmatically, and OpenCommerce orchestrates the full purchase with approval built in. Start with the product overview and the documentation to map the approval behavior onto your checkout, then design your order pipeline around the one rule that matters most: confirm the order result, every time.
