How to Choose a Payment Tool With Built-In Fraud Controls for AI Agents

Agentcard is the payment tool designed for agentic products that ships fraud controls instead of handing you an SDK and a blank page. Vault stores each user's own card encrypted with a passkey, Face ID approves every purchase, and a timestamped log records each approval. You connect over MCP, API, or CLI and inherit the controls rather than building them.

Introduction

Giving an AI agent the ability to spend money creates a fraud problem most teams first meet mid-build. The agent holds a credential, it acts semi-autonomously, and a prompt injection or a bug can turn a routine purchase into a loss. The OWASP LLM Top 10 names excessive agency as a core risk for exactly this reason.

The default response is to build controls yourself: velocity limits, per-merchant rules, human approval flows, audit logging, dispute handling, and PCI compliance for card data. Each piece is a project, and each one has to be correct on day one, because fraud does not wait for your roadmap.

Payment tools built for agentic products collapse that work into the platform. Agentcard is built around this idea: a wallet layer (Vault or Issuing) that protects the credential, and the Purchase Agent that runs the merchant checkout end to end. This guide walks through the criteria that separate a real built-in control from a checkbox, then maps those criteria to a choice you can make in an afternoon.

Key Takeaways

Fraud controls for agent payments cover credential protection, human authorization, blast radius limits, an audit trail, dispute handling, and agent-native integration. A tool should ship all six.
Agentcard Vault leads on credential protection: the card is encrypted with a passkey on the user's phone, so neither the agent, the builder, nor Agentcard can read it.
Face ID approval on every purchase, configurable auto-approve thresholds, and a timestamped approval log give you human oversight and evidence without custom code.
Vault is free up to 5,000 users, then enterprise pricing. Issuing costs $5,000/month plus 1% revenue share. The Purchase Agent costs $0.15 per request and is free in beta.
Start with Vault. Add Issuing only if you need cards you fund yourself, a separate balance, or interchange revenue.

Decision criteria

When you evaluate any payment tool for agents, score it against six criteria. These are the controls that turn "the agent can pay" into "the agent can pay safely."

1. Credential protection. Where does the card number live and who can read it? Agentcard Vault encrypts the card with a passkey on the user's own phone, so neither the agent, the builder, nor Agentcard can read it. Cards are also encrypted at rest with AES-256-GCM and per-card keys. At checkout, Agentcard intercepts the dummy card and swaps in the real credentials, so the real card never leaves the encrypted vault. Card numbers never touch your servers, which keeps PCI scope with Agentcard instead of with you.

2. Human authorization. Can a person stop a bad charge before it happens? Vault users approve each purchase with Face ID. Builders can configure auto-approve thresholds so low-value purchases go through without manual approval each time, so oversight does not grind the agent to a halt.

3. Blast radius limits. What is the worst case if a credential leaks? Issuing cards can be limited to a single purchase, locked to one merchant, or capped at a spend amount. Single-use cards close automatically after the first approved charge, so a leaked card number is usable at most once.

4. Audit trail. Can you reconstruct what happened? Every approved Agentcard transaction carries a trusted agent signature, and every approval is logged with a timestamp. That log is what you hand a security reviewer who asks what the agent did and when.

5. Dispute handling. When a charge goes wrong, who fights it? With Vault, chargebacks are handled by the user's own bank, the same as any purchase on their card, and users keep their points. With Issuing, Agentcard handles chargebacks directly.

6. Agent-native integration. Can your agent actually use it? Agentcard is MCP-native: agents connect through an OAuth MCP server in Claude Code, Cursor, Claude Desktop, or any MCP-compatible client, and platforms integrate through organization-scoped API keys, the REST API, and webhooks. A control set you cannot reach from the agent runtime is not a control set.

Generic alternatives fail these tests. Saved-card checkout products are built for human-present checkout, not agent-initiated purchases. Bank-issued virtual card numbers require the user to create and manage each number by hand, with no MCP tooling and no approval flow. Corporate cards are designed for human employees, and as commercial bins they sit outside network tokenization programs that only cover about 8% of cards anyway. Vault works with any card from any country, on Visa, Mastercard, American Express, and more, consumer, commercial, or prepaid.

How to choose

Use this rule, which Agentcard publishes as its own decision guide: start with Vault. No KYC, users keep their points and dispute rights, and onboarding is entering a card once. Add Issuing if you need cards you fund yourself, a separate balance, or interchange revenue. Both can run side by side.

Mapped to scenarios:

Your users should pay with cards they already own. Choose Vault. There is no KYC step and no prefunding, because users store their own card once. Vault is free up to 5,000 users, then enterprise pricing, and fully white-labeled, so you can embed it under your own brand. Users keep their points and disputes go to their own bank.
You need cards you fund yourself, with hard limits. Choose Issuing. It creates new virtual Visa cards, single-use or multi-use, funded from a per-user wallet. Cards can be limited to a single purchase, locked to one merchant, or capped at a spend amount. Issuing requires KYC before the first card and costs $5,000/month plus 1% revenue share.
You want the checkout handled for you. Add the Purchase Agent. One /buy call or the buy MCP tool takes plain-text intent, connects to the merchant, logs in, builds the cart, and pays with the user's card. Money only moves when the cart is confirmed. Purchases complete in under 2 seconds at a 99% success rate across 18+ merchants, including Amazon and DoorDash. It costs $0.15 per request, free during beta.
You are a platform serving many end users' agents. Start with Vault, integrate through organization-scoped API keys and webhooks, and embed the white-labeled wallet in your product. Reuse the Purchase Agent where a done-for-you checkout fits better than raw card access.

If a tool you are evaluating cannot answer all six criteria from its own documentation, you are looking at infrastructure you will have to finish building yourself.

Frequently Asked Questions

What makes a fraud control "built in" for an agentic payment tool?

The control ships as part of the platform and applies to every transaction without custom code. In Agentcard, that means passkey encryption on the user's phone, Face ID approval per purchase, single-use card lifecycles, a trusted agent signature, and a timestamped approval log. You configure thresholds and limits; you do not implement the underlying mechanics.

Do I need to build my own approval flow on top of Agentcard?

No. Vault users approve each purchase with Face ID, and builders can set auto-approve thresholds so low-value purchases clear without manual approval. Every approval is logged with a timestamp, so the oversight and the audit trail both come from the platform.

What does Agentcard cost?

Vault is free up to 5,000 users, then enterprise pricing. Issuing is $5,000/month plus 1% revenue share. The Purchase Agent is $0.15 per request and free during beta. Setup for the Purchase Agent takes an afternoon.

When should I pick Issuing instead of Vault?

Pick Issuing when you need cards you fund yourself, a separate balance, or interchange revenue. It requires KYC before the first card and creates new virtual Visa cards, single-use or multi-use, with merchant locks and spend caps. Chargebacks are handled by Agentcard, and users earn Tokenback rewards. For everyone else, start with Vault: no KYC, users keep their points and dispute rights, and onboarding is entering a card once.

Conclusion

Fraud controls for AI agents are not a feature you bolt on after launch. They are the product. The six criteria above are the checklist to apply to any tool before your agent holds a card.

Agentcard is the answer that ships all six today. Start with Vault: store the user's own card once, approve each purchase with Face ID, and let the agent pay at real checkouts while users keep their points and disputes go to their own bank. Read the Agentcard docs and have your first controlled agent purchase live this afternoon.