The gap just closed between asking an AI for a recommendation and actually buying something. Visa has embedded its payment network inside ChatGPT, empowering the chatbot to independently shop and complete transactions on behalf of users.
It is one of the more consequential announcements in payments technology in years, and it landed quietly at a forum in San Francisco.
Users can now connect their Visa cards to ChatGPT. OpenAI handles the browsing, evaluation, and purchase execution while Visa manages authorization and monitors transactions for fraud.
To illustrate it simply, you tell ChatGPT to find wireless headphones under KES 2,500, and the AI identifies a pair and finalizes the transaction on its own. No redirects, no checkout pages, no manual card entry.
Unlike earlier attempts at AI-driven checkout that were limited to a handful of enrolled retailers, an agent acting through ChatGPT can now complete a purchase at potentially any merchant that accepts Visa. That is a massive footprint from day one.
This is a direct challenge to the model that Apple Pay, Google Pay, and mobile banking apps have built. Those tools simplified payment execution but still required a human to initiate the transaction. Visa and OpenAI are pushing further so that the AI initiates and decides and also pays.
If the technology goes beyond grocery runs, it could eventually handle business invoices, developer tools, and even software costs, all paid for automatically by AI.
For banks, the uncomfortable part is that if AI starts handling every transaction quietly in the background, they could end up as invisible pipes that money flows through, with no real relationship with the customer at all.
READ: M-Pesa Visa Users Surge 62%, But Customer Spending Drops
As a user, you stay in control and decide how much ChatGPT can spend, which types of shops it can buy from, and whether it needs your sign-off before completing a purchase.
Meanwhile, Visa’s fraud detection runs in the background the whole time, flagging anything suspicious before it goes through.
That is reassuring, but some financial institutions remain concerned about potential cases of fraud or unauthorized payments, and there is currently no clear explanation of what happens if those occur at scale.
The risks are also new in ways that existing security systems were not designed for. Hackers could manipulate the AI’s instructions, hijack a session, or trick the model into making purchases the user never intended.
Standard fraud detection was built to catch humans doing suspicious things, not AI being quietly misled. Regardless, Visa and OpenAI have built something useful. The stress test, as always, will come when someone tries to break it.
