How are agent payments different from card-on-file?

Card-on-file is a standing credential that any system with access can use for any amount. An agent payment uses a scoped credential issued just for an agent session: bound to a merchant (or merchant set), a max amount, an expiry, and an authorized scope. If the agent goes off-track, the worst-case spend is bounded by the token, not the underlying card limit.

Who issues agentic tokens?

The card networks (Visa, Mastercard) and PSPs that have stood up agent-payment programs (Stripe Issuing, PayPal). The token is issued against the user's underlying card or wallet, scoped at the time the agent requests authority. The mechanics resemble network tokenization for normal e-commerce, with the new fields being agent identity, scope, and authority window.

How are disputes handled?

The agent-payment protocols (Visa Trusted Agent Protocol, Mastercard Agent Pay) preserve the chain from transaction → agent → user, so issuers can distinguish a properly-authorized agent purchase from an unauthorized one. Disputes follow the existing card-network rails, with the agent's attestation as additional evidence — closer to a 3-D Secure-style flow than a chargeback-as-default model.

Agent Payments

Written by Max Zeshut

Founder at Agentmelt · Last updated Jul 8, 2026

Payment flows initiated by an AI agent on behalf of a user or business buyer, using a scoped credential issued for that agent's session (an 'agentic token') rather than the user's standing card-on-file. The credential carries the authority limits — merchant allow-list, max amount, expiry, scope — so the issuer, network, and merchant can each apply agent-specific risk and dispute logic. The 2025 protocol foundations are Visa's AI-ready cards (under Intelligent Commerce), Mastercard's Agentic Tokens (under Agent Pay), and Stripe Issuing's agent-scoped virtual cards used by the Stripe × OpenAI [[agentic-commerce]] protocol. Agent payments are the *credential layer* of agentic commerce; [[delegated-purchasing]] is the authority pattern above them.

Example

A user authorizes their AI assistant to handle weekly grocery reorders up to $200. The assistant requests a Mastercard Agentic Token scoped to a specific merchant, capped at $200, expiring after the order completes. The merchant sees an Agent Pay-attested transaction and processes it on the normal card rail; if the assistant tries to order from a different merchant or exceeds the cap, the token is rejected at the network — not at the merchant — preventing scope creep.

Frequently asked questions

How are agent payments different from card-on-file?: Card-on-file is a standing credential that any system with access can use for any amount. An agent payment uses a scoped credential issued just for an agent session: bound to a merchant (or merchant set), a max amount, an expiry, and an authorized scope. If the agent goes off-track, the worst-case spend is bounded by the token, not the underlying card limit.
Who issues agentic tokens?: The card networks (Visa, Mastercard) and PSPs that have stood up agent-payment programs (Stripe Issuing, PayPal). The token is issued against the user's underlying card or wallet, scoped at the time the agent requests authority. The mechanics resemble network tokenization for normal e-commerce, with the new fields being agent identity, scope, and authority window.
How are disputes handled?: The agent-payment protocols (Visa Trusted Agent Protocol, Mastercard Agent Pay) preserve the chain from transaction → agent → user, so issuers can distinguish a properly-authorized agent purchase from an unauthorized one. Disputes follow the existing card-network rails, with the agent's attestation as additional evidence — closer to a 3-D Secure-style flow than a chargeback-as-default model.

Related glossary terms

Related niches

Back to glossary

Loading…