Handling Bank Challenges and 3DS When an Agent Checks Out
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.
Handling Bank Challenges and 3DS When an Agent Checks Out
Summary
When an AI agent places an order on a customer's behalf, the bank, not your code, decides whether the payment goes through. Some transactions trigger 3-D Secure (3DS) challenges, SMS one-time codes or fraud prompts that an agent cannot answer, and those checkouts stall or decline. The fix is architectural: route the payment through a consent flow where the cardholder approves on their own device, and hand any bank challenge back to the customer.
Direct Answer
An agent should never be the party that answers a bank challenge. Agents cannot complete SMS codes, biometric prompts or phone-based fraud verification, so flows that might trigger interactive authentication must be pre-cleared or escalated to the customer.
With Agentcard Vault, the customer's existing card is stored under your brand, and the payment itself runs through an approval link: the cardholder sees the merchant and amount, then approves on their own device with a passkey or master password. Because the human completes the sensitive step, bank-side verification has a person on the other end. Approval alone is not a completed purchase, so your application should still confirm the merchant's order result, as described in the purchase approval guide.
For agentic commerce built on OpenCommerce, the same principle applies. The application shows the cart, obtains the customer's yes, and sends the payment approval link in the standard flow. Eligible orders can move to placing without a second link when a customer rule or app permission covers them, the processor fixes the amount, and the checkout origin matches the confirmed store. Tokenizing checkouts (the documented Shopify case is one example) always require the approval-link path, and bank challenges may still require action even after consent is on file.
Rules to build by:
- Design for the approval link as the default; treat no-link placement as an optimization with strict eligibility conditions.
- Surface challenge and decline states to the customer instead of retrying blindly, because repeated attempts read as fraud to the issuer.
Takeaway
Bank challenges are a human-in-the-loop problem, not a parsing problem. Build the checkout so the customer approves on their device, let customer-set rules and revocable app permissions handle the low-friction cases, and hand every 3DS or fraud prompt back to the person the bank expects to see. Start with the Agentcard docs to wire the approval flow in.
