Visa CEO Says Trust Is the Barrier to AI Payments: Why Autonomous Checkout Stalls and What Merchants Should Do
Visa CEO Ryan McInerney says consumers shop with AI but will not let agents pay. What the trust barrier actually is, how Trusted Agent Protocol and rival standards answer it, and what e-commerce merchants should prepare now.
Key Takeaways
- On September 8, 2026, Visa CEO Ryan McInerney told the Goldman Sachs conference that adoption is happening for shopping but not yet for autonomous payments, and that the barrier, in one word, is trust
- In the survey he cited, three quarters of consumers did not trust agentic platforms to pay autonomously with their money, while 61% said they would trust an agent if Visa were involved, rising above 70% among weekly LLM users. Trust is turning into a technical problem of proving identity and authorization
- For merchants the immediate question is not autonomous checkout itself. It is whether your site can recognize a legitimate agent and avoid losing the AI-referred traffic that is already arriving today
What the Visa CEO actually said about where AI commerce stands
Artificial intelligence may be changing where consumers decide what to buy before it changes where they pay for it.
www.pymnts.comSpeaking at the Goldman Sachs Communacopia + Technology Conference on September 8, 2026, Visa CEO Ryan McInerney split the progress of AI commerce into two very different halves. Consumers routinely use large language models and other platforms to compare, discover and identify products. When it comes time to complete the transaction, however, they still move to the seller's website.
We are seeing adoption for shopping, but not yet for autonomous payments. The barrier to that, if I had to describe it in one word, would be trust.Source: Ryan McInerney, CEO of Visa
What makes the remark worth reading closely is who is making it. Visa runs Trusted Agent Protocol and Visa Intelligent Commerce, so it has every incentive to talk the market up. When a party with that incentive separates discovery from payment and says only the first half has landed, the split itself is the signal.
The trust barrier stands on both sides of the checkout
McInerney framed trust as a two-way problem rather than a consumer-side hesitation. Sellers need to know that agents arriving on their sites are legitimate and have been empowered by consumers to transact. Consumers, for their part, have to trust agents with their money and financial information. Until both conditions are met at once, the flow keeps stopping just short of the cart.
The numbers he cited were blunt. Three quarters of surveyed consumers did not trust agentic platforms to make payments autonomously with their money and financial information. Asked whether they would trust an agent to pay if Visa were involved, 61% said yes, and that figure exceeded 70% among consumers who use LLMs at least weekly.
The gap between those figures is the interesting part. What consumers are refusing is not the idea of an AI paying, but the idea of an unfamiliar party reaching into their account. If an established brand and its guarantees standing behind the transaction move the number that far, then the barrier looks less like emotion and more like a design gap in showing where liability sits. The countries surveyed, the sample size and the fielding dates were not disclosed in his remarks.
Independent data is more cautious than Visa's own
Read only the numbers from the companies pushing this forward and adoption looks one step away. Third-party research paints a slower picture.
Forrester principal analyst Chuck Gahun argued in February 2026 that agentic commerce's next phase hinges on closing the consumer trust gap. In Forrester's Consumer Pulse surveys, US consumer adoption of answer engines for product discovery was essentially flat between February and October 2025, moving from 18% to 19%. As of October 2025, only 8% of US online adults had used OpenAI's Instant Checkout. The pace, in that reading, is not set by the existence of protocols.
Research published by Checkout.com in June 2026 points at the same mismatch on the consumer side. In its finding that consumer demand is forming fast while trust is still catching up, 33% of consumers expect at least 10% of their purchases to be AI-driven within a year, yet 24% say they will never delegate purchases to AI and 27% trust no organization at all to operate an AI shopping agent.
| Source | Period | Reported figure |
|---|---|---|
| Visa CEO remarks (survey details not disclosed) | September 2026 | Three quarters of consumers do not trust autonomous agent payments. 61% would trust an agent if Visa were involved, above 70% among weekly LLM users |
| Forrester Consumer Pulse | February to October 2025 | US use of answer engines for product discovery essentially flat, 18% to 19% |
| Forrester (Instant Checkout usage) | October 2025 | Only 8% of US online adults had used OpenAI's Instant Checkout |
| Checkout.com research (six markets incl. UK and US) | June 2026 | 33% expect at least 10% of purchases to be AI-driven within a year, 24% will never delegate purchases to AI, 27% trust no organization to operate an AI shopping agent |
The conditions for delegation are specific. Across the six markets surveyed, consumers would allow an agent to spend an average of 177 pounds per purchase without additional approvals, and their top non-negotiables were spending caps (30%), instant revocation (29%) and easy cancellation (28%). Put plainly, consumers are not asking for a smarter agent so much as a stoppable one.
The industry answer is converging on agent identity
Replacing trust with verification is already taking concrete shape. In October 2025 Visa published Trusted Agent Protocol, developed with Cloudflare. It gives an agent a cryptographic way to prove its identity and its consumer-granted authorization directly to a merchant, with signatures carrying timestamps, session identifiers and key identifiers so that relays and replays fail. At launch Visa noted that AI-driven traffic to US retail sites had surged more than 4,700% over the prior year. We covered the specification in our explainer on Trusted Agent Protocol.
No single company owns the answer. Mastercard is binding authorization scope into agentic tokens through Agent Pay, while OpenAI's ACP and Google's UCP standardize the commerce flow itself. Visa's own Intelligent Commerce Connect, announced in April 2026, lets a business accept payments initiated through Trusted Agent Protocol, ACP, UCP and Machine Payments Protocol from a single integration. For merchants, being ready for whichever protocol arrives matters more than picking a winner.
Fraud defense moves upstream from the transaction to identity
McInerney also described a shift in where the defensive line sits. Visa has traditionally given banks and merchants tools to identify fraud at the moment of the transaction, but clients now want products that address identity risk before it becomes a fraudulent transaction. Calling identity a critical area of vulnerability, he placed the BioCatch acquisition announced on August 3, 2026 in that frame. The 2.4 billion dollar cash deal is aimed at protecting identities on mobile devices before identity theft turns into a fraudulent payment.
Once agents initiate transactions, the decisive question is no longer whether the card credential is valid but whether the person behind the instruction is genuine. Transaction-time anomaly detection cannot catch a fraudulent instruction executed correctly on a legitimate card.
What merchants should prepare now
The absence of autonomous checkout is not a reason to wait. If McInerney's split is right, the traffic has already arrived even though the payments have not. Shoppers are comparing in an AI surface and landing on merchant sites today, and losses are happening there.
Three items deserve priority. First, check whether bot detection is mistakenly blocking legitimate agent-driven visits, which Visa itself listed as the leading operational problem. Second, decide whether to implement Trusted Agent Protocol, ACP or UCP directly or to wait for your processor or platform, since a processor that implements once covers its whole merchant base. Third, settle liability for returns and mis-orders: when an agent orders the wrong item from inaccurate product content, who processes the return is a question Forrester lists as unresolved.
Product data work pays off immediately. Forrester attributes part of the flat answer-engine adoption to the absence of comprehensive, accurate product data. Information an AI can parse correctly decides whether you make the shortlist long before autonomous payment becomes routine.
Conclusion
The Visa CEO's line that shopping has arrived while payments have not is a statement about sequence, not a dismissal of agentic commerce. If trust is the barrier, what breaks it is not consumers growing accustomed to the idea but machine-verifiable agent identity and authorization, paired with spending caps and instant revocation that let people stop an agent.
Watch two things next: how far identity verification such as Trusted Agent Protocol spreads to merchants through processors, and whether the share of autonomous payments finally starts moving. When it does, merchants who have not prepared the receiving side will be chosen at the comparison stage and lost at the last step.
