June 15, 2026

Visa Wires Itself Into ChatGPT, Mastercard Lets Machines Pay Each Other, the Trust Gap Holds

June 9-15, 2026. June 10 was the busiest single day agentic commerce has had. Visa and Mastercard each shipped their agentic payment rails within hours of each other, Europe’s retail body put agentic commerce on the main stage, and a new consumer study explained why none of it has translated into mass adoption yet.

Last week we covered Britain’s first in-app agentic checkout, OpenAI turning Codex into a store builder, and Mastercard folding agentic payments into its core org. That last move was the setup. This week was the payload. On June 10, at the Visa Payments Forum in San Francisco, Visa announced it will embed Visa payments across OpenAI’s products, and rolled out a suite of merchant-facing agentic tools. The same day, Mastercard launched Agent Pay for Machines, a protocol that lets AI agents pay each other directly, down to fractions of a cent. Also on June 10, EuroCommerce put agentic commerce on the agenda at its Retail Innovation Summit, and checkout.com published research showing the demand for AI shopping is real but the trust to act on it is not there yet.

Here’s what happened.

Visa plugs its network into ChatGPT and Codex

On June 10, Visa announced a partnership with OpenAI to power agent-initiated payments across OpenAI’s products. The mechanics are the part to read. When a user grants permission, an AI agent inside OpenAI’s tools will be able to pay at, in principle, any of the more than 175 million merchant locations that accept Visa, using tokenized Visa credentials so card details are never handed to the agent. Transactions run inside user-set guardrails: spending limits, merchant-category restrictions, and required approvals, with real-time fraud monitoring on top. The integration spans both the consumer side (payments inside ChatGPT) and the developer side (payment acceptance built with Codex), so the same merchant could be sold to by an agent and could build its own agent-payment flow on the same rail. Visa’s Jack Forestell framed the scale of the bet plainly: “AI will transform commerce more profoundly than the internet or mobile technology ever did.” OpenAI’s commerce partnerships lead, Marco Mahrus, put it in terms of where buying is headed: “Commerce is going to happen in many more places and in many more ways than it does today, and agents will play an increasingly important role in helping people complete tasks that involve money, from purchases and payments to more complex transactions.”

The OpenAI deal headlined a wider Visa announcement. At the same forum, Visa unveiled a set of agentic-commerce tools that matter more to merchants than the OpenAI logo does. Agent Score rates a merchant’s website on whether AI agents can navigate it, understand it, and complete a task on it. The Agentic Directory is a registry of agents and merchants Visa has verified as legitimate, so the two sides can trust each other before money moves. And a Large Transaction Model, trained on billions of transactions, is aimed at catching agent-driven fraud while cutting the false declines that would otherwise kill a legitimate agent’s purchase.

Why it matters for merchants: Two things, and the smaller announcement is the bigger one for you. First, the OpenAI deal removes the last excuse for treating ChatGPT shopping as a single-retailer demo. The previous Instant Checkout experiment routed to a handful of partners; a Visa rail means an agent can in theory pay any Visa-accepting merchant, which is most of them. The checkout barrier is effectively gone, which throws the whole weight back onto discoverability, whether the agent finds and chooses you in the first place. Second, Agent Score is the tell. A global card network now thinks “can an AI agent read and use your website” is a number worth publishing, the way a credit score or a Lighthouse performance score is. That is the discoverability problem we have hammered since the start, formalized into a metric a network will hold you to. The action is not to wait for your Agent Score to arrive. It is to make sure that when an agent lands on your product page, it finds complete attributes, clear pricing, real inventory, and answers to the questions a shopper asks, because that is what any score like this will measure. (See our coverage of the 99% catalog-readiness gap for how far most feeds are from passing.)

Mastercard launches Agent Pay for Machines

Also on June 10, Mastercard launched Agent Pay for Machines, an open protocol that lets AI agents send and settle payments to each other autonomously, including micropayments worth fractions of a cent. This is a different animal from the consumer Agent Pay we have tracked, the credential layer that lets an agent buy on a shopper’s behalf. Agent Pay for Machines is machine-to-machine: one agent paying another for a service, an API call, a data feed, or a unit of compute, at machine speed and machine scale, without a human in the loop on each transaction. Per Fortune’s reporting, more than 30 companies signed on at launch, and the roster is the message: card and processing players (Stripe, Adyen, Checkout.com, Global Payments) sitting next to crypto and stablecoin infrastructure (Coinbase, Polygon, Solana Foundation, RippleX, OKX, Ant International). Last week Mastercard moved stablecoins and agentic payments into the same org as the card. This week it shipped the protocol that uses all three.

Why it matters for merchants: This one is not a checkout you will integrate next quarter, and it would be easy to file it under “machine-to-machine, not my problem.” That would be a mistake. Agent Pay for Machines is the economic plumbing for a world where agents do not just buy your product, they buy services from each other on the way to buying your product: the agent pays for a price-comparison call, a fraud check, a delivery quote, a piece of compute, and settles each one in sub-cent increments. The strategic read is that the cost of an agent doing work is dropping toward zero and getting metered by the transaction, which means more of the buying journey will be brokered by agents talking to other agents before a human ever sees a result. The partner list also tells you the settlement layer underneath agentic commerce is being built to be card-agnostic and currency-agnostic, with stablecoins as a first-class option, not a fringe one. The merchant takeaway is the same as on the Mastercard side last week: build to the protocol layer, not to one provider, so you inherit these rails as they switch on. (See our UCP vs ACP breakdown for the protocol map this settles into.)

Update: the following week, Shopify made building on that protocol layer fully self-serve and Adyen shipped a “universal translator” for it. See our coverage of the tooling layer arriving.

The same week, a study explains why none of this is mainstream yet

While the networks shipped rails, checkout.com published research on the gap between consumer demand and consumer trust. The demand is real: a third of consumers (33%) expect at least 10% of their purchases to be AI-driven within a year, with saving time and never missing a better deal cited as the top reasons. The trust is not there to match it: 27% say they trust no organization at all to run an AI shopping agent, and 24% say they will never delegate a purchase to AI. Willingness splits hard by category. Consumers will hand over groceries (41%) and household supplies (31%), the low-stakes repeat buys, but only 15% will let an agent touch financial services. And the merchant side has its own gap: 72% of merchants agree consumers will adopt agent-led shopping faster than most merchants are prepared for.

Why it matters for merchants: This is the corrective to a week of plumbing announcements. The rails are arriving faster than the willingness to use them, and the shape of that willingness is the useful part. Adoption is going to come category by category, starting with the boring, repeatable, low-consideration purchases, not the high-margin considered ones. If you sell groceries, consumables, or replenishable household goods, the agent is coming for your category first, and the trust to use it is already the highest there. If you sell considered or high-ticket items, you have more runway, but the 72% merchant figure is the warning: the people who sell think this lands before they are ready. The move is to get the unglamorous parts right now, the clean catalog and accurate availability that let an agent transact a low-risk repeat buy without friction, because that is the wedge where adoption starts. The trust gap is not a reason to wait. It is a map of which of your products go first.

Update: by July, the gap had started to close. Prime Day data showed AI-referred shoppers converting better than every other channel, a reversal from a year earlier.

Europe puts agentic commerce on the main stage

The same day, June 10, EuroCommerce held its Retail Innovation Summit, with a panel on agentic commerce and the launch of a joint EuroCommerce and McKinsey AI report. The backdrop is the regulatory clarity European retailers have been asking for. Under the EU’s Digital Omnibus agreement on AI, the strict timelines for high-risk AI systems were pushed back (to December 2027 for standalone systems and August 2028 for embedded AI), and the interaction between the AI Act and existing sectoral law was clarified. EuroCommerce, the body representing European retail and wholesale, welcomed the move while pressing for “clear, workable and proportionate rules” applied consistently across the Single Market.

Why it matters for merchants: For European sellers, the signal is that agentic commerce has graduated from a US import to a board-level topic for the continent’s own retail establishment, with a McKinsey report to wave at the leadership team. The regulatory read is the more practical one. The delayed high-risk timelines mean the EU AI Act is less of an immediate brake on agentic experimentation than the August 2026 date suggested, which buys European merchants room to test agentic discovery without waiting for the full rulebook to land. That said, the AI transparency obligations (labeling AI-generated content, disclosing agent interactions) are still coming, so the right posture is to experiment now on the discoverability side, where there is little regulatory exposure, while keeping the disclosure rules in view for anything customer-facing. The European window we flagged in April is still open, and this week it got both a McKinsey blessing and a little more time. (See our coverage of Europe’s first-party channel and regulatory lag for the timeline this extends.)

Update: The transparency obligations were the part that did not slip. Article 50 applies from 2 August 2026, with the content-marking half deferred to 2 December for systems already live. We later covered the August 2 date and EuroCommerce’s lobbying to exempt AI-generated product ads, then the split deadline.

Where the AI commerce stack stands this week

Layer	Update this week
Agent payments (consumer)	Visa to embed Visa payments across OpenAI’s ChatGPT and Codex, 175M+ merchant reach
Merchant readiness	Visa launches Agent Score (rates if agents can use your site) and an Agentic Directory
Agent payments (machine)	Mastercard ships Agent Pay for Machines, sub-cent agent-to-agent payments, 30+ partners
Consumer trust	checkout.com: 33% expect 10%+ AI-driven buys in a year, but 24% will never delegate
Europe	EuroCommerce Summit puts agentic commerce on stage, EU AI Act high-risk timelines slip

What merchants should do this week

1. Treat “can an agent use my site” as a metric you will be scored on. Visa’s Agent Score makes the discoverability problem official: a card network is now rating whether agents can read and transact on your storefront. Do not wait for the score. Audit your product pages the way an agent would: complete attributes, accurate pricing, real-time inventory, and direct answers to the questions a shopper asks. That is what any agentic-readiness score will measure.

2. If you sell groceries, consumables, or repeat-buy goods, the agent is coming for your category first. The trust data is unambiguous: low-consideration, repeatable purchases are where consumers will delegate first (41% for groceries). Get the unglamorous basics right now, clean catalog and accurate availability, so an agent can complete a low-risk repeat buy without friction. That is the wedge where adoption starts.

3. Build to the protocol layer, not to one provider. Both networks shipped rails this week, Visa on the consumer side, Mastercard for machine-to-machine, and the settlement layer underneath is being built to be card-agnostic, currency-agnostic, and stablecoin-ready. The merchant who connects to UCP or ACP inherits these options as they turn on. The one who integrates a single provider pays for each and lags on all of them.

4. European merchants: the regulatory clock just slipped, so use the room. The EU AI Act’s high-risk timelines moved out to 2027 and 2028, which lowers the immediate compliance risk of experimenting with agentic discovery. Test now on the discoverability side, where exposure is low, and keep the AI transparency and disclosure rules in view for anything customer-facing.