The Green Sheet Online Edition
October 27, 2025 • 25:10:02
Insider's report on payments Addressing the downside of AI
Gone are the days when changes in shopping and payment trends were slow to take shape. It took generations from the onset of electronic funds transfers to billions of checks written each year by Americans.
Early in my career the treasurer of one of the largest grocery chains in Florida told me his stores cashed more checks than the largest bank in that state at the time. Recently, that same grocery chain, Kroger, signed an agreement with electronic payment processing giant Worldpay for comprehensive payment processing across in-store, online and mobile shopping channels.
"As payments methods continue to evolve, we look forward to helping keep Kroger at the leading edge of commerce innovation," Jason Pavona, general manager, North America, at Worldpay, said in a press release. What do you want to bet that check clearing isn't mentioned in that agreement?
Meanwhile, it took one generation for ecommerce to become a go-to way to shop. Americans spent $1.34 trillion online in 2024, according to the banking giant Capital One. In the first quarter of 2025, ecommerce accounted for 15.9 percent of all retail sales in the United States, to total $275.8 billion.
AI-driven traffic at retail websites surging
Now, as we approach the year-end holiday shopping season artificial intelligence is coming of age as a tool helping shoppers select that perfect gift. Adobe Research reported that AI-driven traffic to retail websites in the United States surged 4,700 percent over the past year, and 85 percent of shoppers who have used AI to shop said it improved their shopping experience.
According to UserTesting, a provider of human insights for enterprises, one in three consumers across the United States, the U.K. and Australia plan to integrate AI into their holiday shopping this year. Adoption is likely to be highest among Gen Z (56 percent of that generation surveyed plan to use AI) and millennials (50 percent).
The top ways shoppers in the study plan to use AI include:
- Brainstorming gift ideas (39 percent)
- Finding the lowest prices (37 percent)
- Getting direct product links (36 percent)
- Checking whether a deal is truly worthwhile (30 percent)
"Think of AI as the holiday sidekick we didn't know we needed," said Bobby Meixner, vice president of solution marketing at UserTesting. "It's giving shoppers fresh ideas, uncovering smarter deals, and helping them put more meaning into every gift. That's why younger generations are leaning in; it makes tradition more fun and thoughtful."
But is it safe?
Every new iteration of shopping and payments creates new opportunities for fraud. Check fraud, for example, continues to cause billions of dollars in losses for businesses and banks. Fraud takes several forms in ecommerce, the most notable is euphemistically called "friendly fraud," because it usually involves legitimate uses of a card for which the cardholder denies responsibility.
UserTesting's survey suggests consumers trust AI tools as much, or more than, friends and family for gifting advice. But what about AI grifting? How big a concern is that?
"As fraudsters' use of AI tools accelerates, the scope and scale of authorized payment scams edge close to a tipping point that threatens to upend the trust relationship across the whole financial system," said Trace Fooshee, strategic adviser at Datos Insights.
A 2024 poll of U.S. consumers by the cloud security company Upwind found 87 percent of participants worry that AI is helping scammers. Fifty-five percent said their worry about scams is growing. Eighty-six percent expressed increased worry about online scams during the holiday shopping season.
Recently, a report by identity and fraud prevention platform Alloy found 85 percent of Americans in the study worry that AI makes scams harder to detect. To be sure, AI can be used to create realistic product images, payment portals, email links, deepfake videos, and even mimic familiar voices to lure victims into clicking malicious links.
Among the 2,000 Americans surveyed by Harris Poll on behalf of Alloy, 62 percent reported being victims of a scam or knowing someone who had been victimized. Twenty-eight percent reported losing money as a result of an AI-related scam, and of those victims, better 21 percent reported losing $5,000 or more.
"The data confirms what we're seeing on the ground: AI hasn't made fraudsters more sophisticated; it's made them more efficient," said Sara Seguin, principal adviser on fraud and identity risk at Alloy. "A single criminal can now launch thousands of personalized attacks in minutes. But here's what is fascinating: consumers get it. They're demanding banks use the same AI technologies to protect them."
Sixty-seven percent of those surveyed said financial institutions should reimburse money lost to these scams even in cases where transactions were personally authorized. Gen Z and millennials were even more likely to expect reimbursement from their FIs: 73 percent of those generations expected reimbursement compared with 62 percent of Gen X and baby boomers.
AI and payments: card brands enter the fray
Pablo Fourez, chief digital officer at Mastercard, addressed the need to scale AI (specifically agentic AI, or agentic commerce) with trust in a recent blog post announcing the Mastercard agent pay acceptance framework.
Fourez pointed out that with agentic AI proliferating, merchants are increasingly confronted with several critical dilemmas. These include: How can they distinguish between legitimate AI agents and malicious bots? How do they know the consumer authorized the agent to make the purchase? How can they know the agent has carried out the consumer's instructions correctly?
"Without a standardized method to verify the identity of agents and consumer intent, merchants risk increased fraud, reputational damage and ultimately a poor experience for their genuine customers," he wrote. The Mastercard framework aims to "establish an essential consistent standard for agent verification and data exchange," which the card brand is sharing with developers to support "ongoing refinement," he added.
The framework focuses on registering and enabling AI agents to transact with agentic tokens. "Each agent is uniquely identified and then enabled to initiate transactions using agentic tokens," Fourez noted. "This provides the core foundation for all participants to recognize when an agent is involved."
Visa, too, is on the case. It unveiled a trusted agent protocol that enables approved agents to securely pass critical information to merchants. The framework for recognizing trusted agents was created with feedback from leading payment services providers, including Elavon, Fiserv, Nuvei, Stripe and Worldpay.
Visa Chief Product and Strategy Officer Jack Forestell said the card company has also been working with sellers and card issuers on the project. "We believe the entire payments ecosystem has a responsibility to ensure sellers can trust AI agents as much as they trust their best customers and networks," he said in a press release.
Mastercard also has a litany of partnerships tied to its agentic AI vision, Fourez stated. For example, it has partnered with the network services and cybersecurity firm Cloudflare to support web bot authorization to verify AI agent identities. "These partnerships are instrumental in shaping a future where agentic commerce is not only possible, but trusted and ubiquitous," he added.
Coming to terms with the coding 'tax'
In the end, it all comes down to the customer experience, said Peter Dougherty, president of Spreedly, an open payments platform company. "Having a great checkout experience is hard," Dougherty said during a recent interview. "Every new piece you add to that [for example, new payment options] comes with its own basket of complexities."
One complexity is what Dougherty referred to as a "growing engineering tax," which is time spent maintaining evolving checkout flows. A meaningful share of companies reported dedicating 25 percent or more of engineering resources to payments upkeep. "AI should be layered thoughtfully to strengthen these payment systems, not replace them entirely and introduce new risks," he said.
That's where Spreedly can step in, maintaining connectivity to payment services providers, AI agents, etc., Dougherty said.
Forestell, meanwhile, said Visa's new agent protocol "is focused on creating no-code functionality for merchants to securely identify agents with an intent to buy and provide a better payments and personalized experience for its known users."
But what about one-off problems that aren't necessarily tied to payments. Say, for example, I tell the AI agent to find me a good deal on Merrill sneakers, size 10.5, but what gets delivered is a pair size 10. Who owns that mistake? Is it Merrill, is it the merchant website where the purchase was made? "Actually, there's been some news recently on different protocols trying to establish that chain of accountability," Dougherty said. "These are things that all merchants, the merchant ecosystem need to get used to."
Research recently sponsored by Spreedly found executives at middle market retailers fear AI may worsen checkout failures, despite the fact that 83 percent of executives polled expect that by 2027, most payments will be initiated, optimized or completed by AI.
Among the risks associated with AI-led payments those executives cited:
- Losing transparency about who the customer is (cited by 23 percent)
- False positives leading to lost sales (20 percent)
- Losing the ability to market directly to consumers (22 percent)
Only 15 percent of executives polled said they had no concerns—a reminder that progress in AI-driven payments must keep pace with vigilance and trust.
Patti Murphy is senior editor at The Green Sheet, president of ProScribes Ink (www.proscribes.net) and self-described payments maven of the fourth estate. Her Today in Payments reports are a regular feature of the Merchant Sales Podcast.
Notice to readers: These are archived articles. Contact information, links and other details may be out of date. We regret any inconvenience.
