Stay updated with ZDNET: Mark us as a preferred source on Google.
ZDNET’s main points
- Visa will safeguard AI-driven payments within OpenAI platforms.
- The security framework for AI-powered transactions is still evolving.
- Both individuals and companies could face potential threats.
AI-powered shopping represents a fast-expanding area of artificial intelligence for both consumers and businesses, a trend highlighted by Google’s recent introduction of Universal Cart at its I/O event in May. This week, Visa and OpenAI strengthened that foundation further — but can AI agents be trusted to handle purchases responsibly?
On Wednesday, the two organizations revealed a collaboration to enable Visa-secured AI agent transactions within OpenAI, effectively “bringing AI-driven commerce into the mainstream,” according to Visa’s announcement. Visa’s Trusted Agent Protocol, along with its additional authorization and security measures, will be integrated into OpenAI platforms such as Atlas and ChatGPT Shopping, enabling developers and sellers to process payments from AI agents.
(Disclosure: Ziff Davis, ZDNET’s parent company, filed a lawsuit against OpenAI in April 2025, claiming that OpenAI violated Ziff Davis copyrights while training and running its AI systems.)
“Transactions function within boundaries established by the consumer or business: spending caps, mandatory approval levels, and other permission tiers that ensure the buyer remains in charge even while an agent carries out the task,” Visa stated.
For everyday users, this means allowing AI agents to handle your product research (with pre-configured personal preferences) and complete routine purchases you’re comfortable automating. For sellers, the concept remains that a smoother purchasing journey across more AI-driven platforms — where consumers are increasingly spending time — attracts new customers.
Also: How small businesses can thrive amid AI shopping: 7 crucial strategies
Visa’s announcement also noted the collaboration would extend to “OpenAI’s growing lineup of AI-powered products,” though it did not clarify any role in the company’s upcoming “superapp” when ZDNET inquired.
“The objective is to make AI-driven commerce more approachable, reliable, and secure for consumers and businesses around the globe,” the company added, pointing to its Intelligent Commerce platform as the basis for the partnership.
A shifting, risk-filled environment
Major financial institutions adopting AI-driven commerce is not unprecedented. OpenAI has been seriously building its AI commerce capabilities since launching Instant Checkout, which leverages the company’s Agentic Commerce Protocol, created with Stripe, to let sellers securely process transactions through ChatGPT.
At that time, ACP joined Google’s Agent Payments Protocol (AP2) and other comparable initiatives aimed at making AI commerce appear more secure and attractive. Also on Wednesday, Mastercard unveiled Agent Pay for Machines, designed to expand and accelerate transactions between AI agents.
Also: I asked ChatGPT Atlas to handle my Walmart shopping – here’s how the AI browser agent performed
Nevertheless, many users are naturally hesitant about surrendering any portion of the buying process to AI agents, which have a history of behaving unpredictably.
ZDNET asked Visa what guidance it would offer consumers who are uneasy about giving payment authority to an AI agent, even within Visa’s framework, and whether the company has a protocol in place for situations where an agent circumvents protections or triggers an unauthorized transaction.
“Visa’s strategy for AI commerce centers on user control, transparency, and security,” the company responded. “Transactions occur within user-specified permissions, including spending limits, merchant categories, and approval requirements, and utilize tokenized credentials, real-time authorization, and fraud detection.”
However, experts remain cautious.
“Even with robust conventional security measures like tokenization and fraud detection, AI-driven payments bring new dangers that current systems weren’t built to handle — moving the challenge from verifying users to ensuring agents operate within intended boundaries and policies,” said Geoff Cairns, principal analyst at Forrester, in an email. “For consumers and businesses, the primary worries would be unauthorized or erroneous transactions, unclear liability, and fraud that outpaces traditional dispute resolution processes.”
Visa stated that for financial institutions and enterprise clients with rigorous security requirements, monitoring agent credentials across interactions helps prevent them from taking unauthorized actions, creating workflow efficiencies without adding risk.
Beyond that, the typical AI challenges also affect AI-driven commerce. As the Guardian reported last week, AI shopping assistants can present fraudulent sites as legitimate stores. While the payment process itself isn’t the issue here, the broader landscape of AI-assisted shopping is still dealing with trust gaps before a buyer even reaches the payment stage.
So are AI-driven payments worthwhile, at least at this point?
“The convenience advantages of AI payments don’t automatically raise risk, but they transfer authentication from direct user engagement to ongoing, risk-based verification, where delegated authorization and ‘on behalf of’ controls become essential to trust,” Cairns explained. “This is a technical domain that is still maturing.”
