Key Takeaways
- On June 10, 2026, Visa announced a strategic partnership with OpenAI at the Visa Payments Forum, integrating Visa's network, tokenization, and fraud monitoring into OpenAI's agents to enable agent-initiated Visa payments
- Transactions run within user-defined controls such as spending limits, merchant categories, and approval requirements, while Visa handles fraud detection, chargebacks, and refunds. Mastercard announced a machine-to-machine payment protocol the same day, sharpening the race for payment infrastructure
- With more than two-thirds of merchants expecting AI agents to initiate at least 10% of e-commerce transactions within three years, e-commerce and booking businesses now face a practical agenda: making their sites agent-ready and re-evaluating payment partners
Visa's Payment Network Connects to ChatGPT
New collaboration brings Visa's global payment network to one of the largest AI platforms and aims to support seamless, secure transactions across commerce environments supported by agents.
investor.visa.comOn June 10, 2026, on stage at the Visa Payments Forum in San Francisco, Visa announced a strategic partnership with OpenAI. According to the official announcement, Visa will provide its global network, credentialing capabilities, and security infrastructure to enable secure Visa payments within OpenAI's agent experiences. For developers and merchants, this means a standardized path for accepting agent-initiated transactions as Visa payments.
The scope goes beyond shopping in ChatGPT. The two companies say they will jointly explore enterprise applications, including developer-focused experiences powered by the coding agent Codex and more automated workflows. Jack Forestell, Visa's Chief Product and Strategy Officer, declared that "AI will transform commerce more profoundly than the internet or mobile technology ever did," framing the infrastructure around a future where agents become active participants in the economy.
The deal sits within Visa Intelligent Commerce, the initiative Visa has been building since April 2025. It is a platform providing the trust, controls, and connectivity AI agents need to discover products and initiate and complete transactions, with pilots already extended to Europe, Latin America, and Asia-Pacific. Connecting to OpenAI, one of the largest AI platforms, lifts that initiative to consumer scale in a single move.
Spending Limits, Merchant Restrictions, Approvals — Designing Controls Before Handing Over the Wallet
The biggest question in delegating payments to an agent is how much control humans retain. According to the announcement, transactions operate only within clearly defined user permissions, policies, and controls. Concretely, there are three levers: spending limits, allowed merchant categories, and approval requirements under defined conditions. Payments use tokenized Visa credentials, with real-time authorization and fraud monitoring running throughout.
Axios reports that Visa will handle fraud detection, chargebacks, and refunds. Users define the agent's boundaries, and when something goes wrong, remediation flows through the same framework as existing card payments. This is Visa's answer to the question the industry keeps asking: who is responsible when things go wrong.
The use cases extend well past consumer shopping. Business invoice payments are in scope, as is a world where AI coding agents autonomously purchase additional inference, APIs, or developer services. Rubail Birwadker, Visa's global head of growth, described where AI's influence on buying stands today.
More than one in five transactions are really being influenced by what users are learning through LLMs
The experience, he explained, could end up feeling similar to shopping with Apple Pay or Shop Pay. At the same time, both companies acknowledge they do not yet know exactly what this will look like for users. The accurate reading is that this is not a finished product launch but a move to lay plumbing that works regardless of what form agent payments eventually take.
Three Months After Retiring Instant Checkout, OpenAI Rebuilds from the Foundation
Commerce inside ChatGPT is not OpenAI's first attempt. Instant Checkout, launched with Stripe in September 2025, aimed for a native purchase experience completed entirely within ChatGPT, but adoption lagged. According to Finextra, OpenAI ended it in March 2026, six months after its debut, pivoting toward retailers building dedicated apps inside ChatGPT.
That history clarifies what this new arrangement means. Having stumbled on the checkout "screen," OpenAI is now re-laying the payment plumbing itself with Visa. Visa, for its part, has a blueprint to extend the integration built with OpenAI to other platforms over time, as Birwadker told Axios. Both sides are taking a second run with the lessons of the first baked in.
Mastercard Moved the Same Day — The Land Grab Among Payment Networks
On the same June 10, Mastercard announced Agent Pay for Machines. It is a protocol built for a world where AI agents pay each other continuously and at high velocity, processing programmatic micropayments of fractions of a cent, with the permissions humans grant their agents recorded on a blockchain (Polygon). More than 30 companies, including Stripe, Adyen, Coinbase, and Cloudflare, signed on as initial partners.
| Item | Visa × OpenAI | Mastercard Agent Pay for Machines |
|---|---|---|
| Announced | June 10, 2026 (Visa Payments Forum) | June 10, 2026 |
| Target domain | Agent payments by consumers and businesses in ChatGPT and other OpenAI experiences | Machine-to-machine payments between AI agents |
| Payment rails | Existing Visa network with tokenized credentials | Across cards, stablecoins and more; permissions recorded on Polygon |
| Use cases | Delegated purchases, business invoice payments, Codex buying APIs and compute | Sub-cent micropayments, pay-per-use APIs and data |
| Initial partners | OpenAI | 30+ companies including Stripe, Adyen, Coinbase, Cloudflare |
The two strategies are a study in contrast. Visa secured one of the largest consumer AI platforms and chose to make agent payments work on existing card rails. Mastercard is laying new rails in machine-to-machine territory where cards do not yet exist. Given that Mastercard's chief executive had recently voiced grave concerns about consumer protection in agentic commerce, attacking from the machine-to-machine side, where no human is in the loop, reads as a consistent choice.
The market is moving too. In S&P Global's 2026 merchant survey (covered by Forbes), 65% of merchants said they are considering adding a new payment partner for agentic commerce, and more than two-thirds expect AI agents to initiate at least 10% of their e-commerce transactions within three years. Infrastructure decisions are being made before transaction volume arrives.
What E-commerce and Booking Businesses Should Take Away
This is not a story that changes tomorrow's revenue. Visa's own CFO, Chris Suh, told Fortune he is "hesitant to lean into the stablecoin and agentic commerce narratives too much," acknowledging that the vast majority of Visa's business today has nothing to do with them. Yet he added that these investments "won't pay off in the next six months, but could over the next six years." There is always a lag between laying infrastructure and monetizing it.
What should businesses prepare during that lag? The companion tools Visa announced the same day offer a hint. Agent Score evaluates whether AI agents can navigate a merchant's site and complete purchases, while the Agentic Directory is a registry where verified agents and merchants can confirm each other. Being a site agents can read and a store agents can trust is becoming a competitive requirement on par with search engine optimization.
Booking businesses in travel and transportation are not bystanders either. The "tasks that involve money" described by Marco Mahrus, OpenAI's head of partnerships for commerce, include not just purchases and payments but more complex transactions — and reservations with changes and cancellations are the textbook case. The experience of delegating a booking to an agent under spending limits and approval requirements runs precisely on the plumbing this partnership laid.
Conclusion
The Visa-OpenAI partnership moves agent payments from concept to plumbing work. Human control stays in place through spending limits and approval requirements, while fraud and chargebacks are absorbed into the existing card framework. It is not flashy, but the significance of the two giants turning the industry's shared belief — that trust design determines the pace of adoption — into implementation is hard to overstate. Once the pipes are in, water tends to flow quickly. For e-commerce and booking businesses, a quiet preparation period has begun: checking whether your site and your payments are visible to the agents.
