Frictionless shopping in 2027 via AI? Not so fast, counters one former PayPal executive.
John Lunn, now founder and CEO of the payment orchestration platform Gr4vy, says the tech hype is outpacing reality and predicts AI shopping agents will trigger a chargeback nightmare for retailers, along with a flood of fraud.
“Bots promise easier buying, but when an AI agent makes an unauthorized purchase, retailers will likely get stuck holding the bill,” he told the E-Commerce Times.
Lunn recalled speaking with someone who said they had asked an AI agent to buy the cheapest airline ticket between two cities.
“The agent came back and said, ‘I bought your ticket, and it cost zero dollars,'” he noted. “The person asked, ‘How on earth did you do that?’ The agent replied, ‘I went on the internet, found a credit card number, and used that.'”
“Now that’s fraud, but who committed it?” he asked. “The consumer could just go, ‘No, it wasn’t me. I didn’t do that.’ And then it’s on the merchant to prove they did.”
“Everybody would love to say, ‘Can we slow down and get this right?’ but that’s not how AI is running,” he continued. “This rock is rolling downhill, and it’s moving faster and faster, and no one is saying that we’re going to be liable for this, which ultimately means the merchant is, because they’re the ones selling the product.”
Who Pays When an AI Agent Gets It Wrong?
Software providers also have questions about who bears the risk. In a report released Tuesday, Wind River Payments found that 52% of software providers surveyed consider current payment rules clear enough for purchases made by AI agents. The rules cover disputes, chargebacks, fraud liability, and compliance.
The Madison, Wis., payment services company surveyed more than 200 independent software vendors and merchants across the U.S. Among software providers, 43% cited uncertainty over who would be liable for fraud, disputes, or errors as a leading concern.
Among merchants, 41% cited disputes and chargebacks as a top concern, 40% cited fraud and misuse, and 39% cited difficulty verifying that a customer had authorized a purchase.
Asked who should bear the cost of chargebacks and fraudulent purchases, 34% of merchants chose the company running the AI agent. Just 8% said the merchant should pay.
Retailers may not be facing a chargeback nightmare, but they can expect a very bumpy first stretch, observed Stewart Watterson, a strategic advisor at Datos Insights, a research and advisory firm in Boston.
“Under network rules published so far, cardholder protections carry over to agent-initiated purchases, and the merchant of record still owns disputes and representment,” he told the E-Commerce Times.
“American Express announced Agent Purchase Protection on April 14, 2026, but it isn’t live, and it requires the cardmember to try a return first,” he said. “Retailers are pushing for a liability shift toward agents, and that debate isn’t settled.”
“Merchants in Datos Insights interviews expect dispute volume to rise as consumers learn the process, then level off,” Watterson said. “A verifiable record of what the shopper authorized gives retailers better evidence than they have in a typical e-commerce dispute.”
When an AI Purchase Leads to a Return
Shopping agents give customers a new reason to dispute a purchase, according to Mike Smith, founder of BuyerGain, a search marketing agency in Nashua, N.H.
Mastercard’s 2026 research with Datos Insights projects that annual chargeback volume worldwide will reach 334 million by 2028.
“Chargeback rules assume a person clicked ‘buy,'” Smith explained. “When an agent places the order, the shopper can honestly say, ‘I authorized the agent, but I never chose that item,’ and the merchant could be forced to take the loss.”
“If a shopper simply does not like what the agent picked, sending it back is the natural step, and a return costs a retailer far less than a chargeback,” he said. “And is not as damaging to their merchant account.”
Smith expects returns to rise slightly in 2027. “Retailers can steer it that way by making returns on agent orders easy, and by logging what the agent was authorized to do and what it ordered, because that record is what wins the disputes that do get filed,” he noted.
“Additionally,” he continued, “if a shopper is using an agent in a way that creates a lot of returns, the merchant could monitor or restrict their returns just as they do for a shopper that returns a lot of items.”
Why Synthetic Identities Can Pass Checkout
Fraud can begin well before checkout. Gr4vy’s Lunn said criminals use AI to create fake shopper identities that look legitimate by the time an order reaches a retailer.
“AI tools have made it incredibly easy for a fraudster to create identities that look very, very legitimate,” he explained.
“By the time they get to checkout,” he continued, “it’s very hard to differentiate a good user from a bad user, because it’s becoming increasingly easy and cheap to create fake identities, so it’s changed the game when it comes to fraud.”
Patrick Harr, CEO of DataVisor, a Mountain View, Calif., company that uses AI to detect fraud and money laundering, explained that most checkout defenses were built to judge one event at a time, asking whether this order, card or person looks legitimate right now.
“Rules, address and card verification, document checks and velocity limits all work that way,” he told the E-Commerce Times. “They remain useful, but a synthetic persona is designed to pass exactly that kind of test.”
“Because the identity is new, it also has no fraud history, no label and no blocklist entry to match against,” he continued. “Static rules widen the gap, since they go stale quickly when attackers adjust on every attempt.”
“We have observed a 650% year-over-year increase in agentic traffic, reflecting the rapid growth of automated and AI-driven activity online,” added Maanas Godugunur, senior director for fraud and identity at LexisNexis Risk Solutions, a global data analytics and services company.
“As this technology becomes more widely used,” he told the E-Commerce Times, “fraudsters can also leverage AI and automation to operate at greater scale and create more convincing, personalized attacks.”
Do Shoppers Trust AI to Buy for Them?
Chip Lupo, a writer and analyst for WalletHub, an online consumer financial information company, predicted AI will play an increasingly important role in retail in 2027, making shopping more convenient but also raising questions about consumer trust, fraud and security.
“While AI agents can help shoppers compare products, find deals and handle routine purchases, consumers are not yet fully comfortable handing over control,” he told the E-Commerce Times. “For example, 56% of people don’t trust AI to help them with their holiday shopping this year.”
He added that retailers face a two-sided challenge: deploying their own AI tools to simplify shopping, while adapting to third-party AI agents that increasingly shop on consumers’ behalf.
“In both cases,” he said, “success will depend on transparency, protecting customer information and keeping human support readily available.”
“The retailers that come out ahead,” he added, “will be those that use AI to save consumers time without sacrificing the trust and personal connections that drive long-term loyalty.”




