Visa Intelligent Commerce is Visa’s program for letting an AI agent pay with your card. Visa introduced it on April 30, 2025: card details are replaced with tokenized credentials an approved agent can use inside the spending limits and conditions you set. In the US, Visa says transactions were live as of April 29, 2026; in Europe, banks and merchants completed live agent purchases on July 2, 2026.

Adoption is moving faster than trust. Visa’s own research, published on September 9, 2026, found that only 23% of US consumers trust generative AI to handle payment transactions on their behalf, while 61% said they would trust Visa to handle agentic transactions.

What Visa Intelligent Commerce is

Visa describes the program as a way to open its payment network to the companies building AI agents, delivered as “a suite of integrated APIs and a commercial partner program” for AI platforms. The April 2025 launch named collaborators including Anthropic, IBM, Microsoft, Mistral AI, OpenAI, Perplexity, Samsung and Stripe.

The label has grown since. By the Visa Payments Forum on June 10, 2026, Visa was calling Visa Intelligent Commerce its platform for agentic commerce, providing “the trust, controls and connectivity needed for AI agents” to discover, initiate and complete transactions. Visa’s Asia Pacific announcement of November 12, 2025 lists the moving parts as “tokenisation, authentication, payment instructions, and transaction signals.”

In practice it is less a product you sign up for than a set of rails your bank, your AI app and the stores you buy from each have to adopt separately.

How Visa Intelligent Commerce works

Agent-specific tokens

The foundation is tokenization. Visa’s launch release describes AI-Ready Cards that replace card details with tokenized digital credentials, which “confirm that a consumer’s chosen agent is allowed to act” on that consumer’s behalf. The agent never holds your card number.

Stripe, which processes these transactions for sellers, defines the credential from the acquiring side: agentic network tokens are “network-issued, secure digital credentials” that let authorized agents initiate payments without exposing card details, and they carry extra information in the authorization message so issuers can make better decisions and manage disputes and fraud.

Visa said in June 2026 that it is enriching token data with more detail on the transaction type, where the token is used and who is making the payment, and adding a token assurance signal scored from a token’s provisioning and behavioral history.

Authentication and passkeys

A token is only as good as the proof that you asked for the purchase. In Europe, Visa says issuer participation runs through Visa Payment Passkeys, and that each transaction is “securely authorised and directly linked to a verified user” and their explicit instruction. Every transaction completed in Visa’s European Agentic Ready program was secured with passkeys, which Visa ties to Strong Customer Authentication compliance there.

Visa’s own July 2026 agentic payments report describes programmable spend controls as emerging across agent payment systems, alongside tokens and passkey authentication on the card side.

Your instructions and limits

Visa has been consistent that the cardholder sets the boundaries. “Each consumer sets the limits, and Visa helps manage the rest,” said Jack Forestell, Visa’s chief product and strategy officer, at the April 2025 launch. The same release says only the consumer can instruct the agent on what to do and when to activate a payment credential.

The OpenAI collaboration announced June 10, 2026 is more specific about the shape of those rules: transactions operate within defined user permissions and controls, “such as spending limits, merchant categories or required approvals.”

Commerce signals

Alongside the payment, Visa receives context. Commerce signals are shared in real time with Visa, which Visa says lets it apply transaction controls and “help to manage disputes.” This is the piece that decides, in the moment, whether an agent-initiated authorization fits the rules attached to your credential.

The launch release describes AI-Powered Personalization as opt-in: consumers share basic Visa spend and purchase insights “with their consent” to improve an agent’s recommendations. Visa frames the sharing as the consumer’s choice, not a default of using an agent.

What you control as a cardholder

The controls Visa has named publicly are the ones above: which agent is approved, spending limits and conditions, merchant categories, and purchases that require your approval before they go through. Europe’s July 2026 release adds that live agent purchases there were “explicitly authorised by the consumer”, with agents acting inside defined user parameters.

What Visa has not published is a cardholder-facing off switch. If you want to know whether a specific card can be used by an agent at all, or how to stop it, ask the issuer: Visa’s issuer readiness program, Agentic Ready, is run with banks, and in Europe Visa says it worked with over 30 European issuers to enable live agent purchases. Your bank, not Visa, is the party that enrolled your card.

Our separate guide on setting spending limits for AI agents covers where those caps actually live today.

Approvals, protection and who pays for a mistake

Visa’s clearest statement on protection is not agent-specific. In its November 20, 2025 post on agentic commerce threats, Visa writes that when consumers pay with Visa on any platform, “they’re protected by our zero-liability guarantee, ensuring they’re never held responsible for unauthorized charges.”

That covers unauthorized charges. It does not settle the harder case: an agent you did authorize, buying something you did not want. Visa’s own July 2026 report is blunt about the gap. “When an agent makes an unauthorized purchase, the responsible party may initially be unclear,” it says, and “liability remains legally ambiguous” across the human who delegated authority, the platform hosting the agent, the model provider and the merchant.

For the mechanics of refunds and chargebacks in that situation, see our piece on who pays when an AI agent buys the wrong thing.

What merchants and small sellers do

Trusted Agent Protocol. Visa published TAP on October 14, 2025, developed with Cloudflare and released through the Visa Developer Center and GitHub. It uses agent-specific cryptographic signatures to carry agent intent, consumer recognition and payment information, with what Visa calls minimal user experience changes on merchant sites. On the engineering side, Visa’s developer documentation puts it simply: “A merchant’s primary task is to validate this signature.” Cloudflare describes the protocol as a way for payment networks to integrate Web Bot Auth, and Akamai added support in December 2025. Our comparison of Visa’s Trusted Agent Protocol and Mastercard Agent Pay goes deeper on the cardholder-facing differences.

Intelligent Commerce Connect. Announced April 8, 2026 as part of the Visa Intelligent Commerce portfolio, this is a single integration via the Visa Acceptance Platform covering payment initiation, tokenization, spend controls and authentication. It accepts payments initiated through Trusted Agent Protocol, Machine Payments Protocol, Agentic Commerce Protocol and Universal Commerce Protocol, and helps merchants make catalogs discoverable inside AI platforms. Visa lists it as in pilot with Aldar, AWS, Diddo, Highnote, Mesh, Payabli and Sumvin, rolling out to more partners during the year, with features marked subject to availability.

Agent Score and the directory. At the June 2026 Payments Forum, Visa introduced Agent Score, created with New Generation, which “allows merchants to evaluate their websites for agentic commerce readiness” (whether an agent can navigate and complete tasks on the site). Visa also described a directory of agents and merchants “that Visa has verified as legitimate participants in agentic commerce,” called the Agentic Directory in that release and the Agent Directory in the European announcement.

Developer tooling. Visa opened a Model Context Protocol server for Visa Intelligent Commerce APIs on September 4, 2025, with the Visa Acceptance Agent Toolkit built on top of it for invoicing and Pay By Link workflows triggered in plain language. Visa says both are available in pilot.

If you already use a processor. Stripe says that any seller already processing payments with Stripe automatically supports these agentic payment methods, which since March 3, 2026 include Visa Intelligent Commerce.

Where Visa Intelligent Commerce is live

Date What Visa announced Source
April 30, 2025 Visa Intelligent Commerce introduced at the Visa Global Product Drop Visa
September 4, 2025 MCP server and Acceptance Agent Toolkit, both in pilot Visa
October 14, 2025 Trusted Agent Protocol published for merchants Visa
November 12, 2025 Asia Pacific expansion; pilots planned “by early 2026” Visa
December 18, 2025 Hundreds of agent-initiated transactions; US pilots in closed beta with Skyfire, Nekuda, PayOS and Ramp; more than 100 partners Visa
March 3, 2026 Stripe processes Visa agentic network tokens for sellers Stripe
March 17, 2026 Agentic Ready launches first in Europe, including the UK Visa
April 8, 2026 Intelligent Commerce Connect, in pilot with select partners Visa
April 29, 2026 US live transactions; Agentic Ready extends to Asia Pacific and Latin America, 20+ partners already live in the UK and Europe Visa
June 10, 2026 Agent Score, the directory, token enhancements; OpenAI collaboration Visa
July 2, 2026 Live agent purchases in Europe with 30+ issuers and merchants including lastminute.com, Frasers, Cleverbridge and BrickDepot Visa
September 9, 2026 Visa Trust Index: 23% of US consumers trust GenAI with payments Visa

Two things are worth separating. Agentic Ready is a testing program for banks: Visa describes it as a way to test agent-initiated payments in controlled, real-world environments using live cards and real merchants. Live transactions in the US and Europe are real, but they run across a named set of agents, issuers and merchants rather than everywhere Visa is accepted.

Open questions Visa itself raises

Visa’s July 2026 report is unusually candid about what is unresolved:

  • Mis-purchase at machine speed. The report notes that a single misconfigured agent can execute hundreds of bad transactions before anyone notices, creating refund and dispute overhead that is hard to contain.
  • Prompt injection. Manipulating an agent through malicious input is, in the report’s words, “an active area of research with no complete solution.”
  • Disputes. “Disputes lack established resolution mechanisms,” it says, because chargeback windows and evidence requirements “were designed for human-speed commerce with clear buyer intent.”
  • Regulation. The report says treatment of agent-initiated transactions, liability and cross-border agent commerce remains unsettled.

Visa’s security team adds a fraud angle. Its November 2025 threat post warns that counterfeit storefronts can be engineered specifically to deceive shopping agents by passing automated checks, undercutting market prices, then harvesting the credential used. Visa reported a 25% rise in malicious bot-initiated transactions over the prior six months, with a 40% rise in the US.

How we checked this

Facts checked on October 20, 2026. Every figure, date and quotation here comes from Visa’s own press releases and developer documentation, Visa’s July 2026 agentic payments report, and partner announcements from Cloudflare and Stripe. Where Visa’s wording for a capability changed between releases, we use the name each release carries.