Banks Warn AI Shopping Agents Are Outpacing Fraud Protections

Imagine handing your credit card to a stranger and saying "buy me something nice, and don't get scammed." That's roughly the leap of faith consumers are being asked to make as AI shopping agents move from novelty to normal. And according to a coalition of major banks, the safety net hasn't caught up.

NatWest, Bank of America, ING, Capital One, New Zealand's ASB Bank and Commonwealth Bank of Australia joined together this week to sound the alarm on what they call agentic commerce: the growing practice of letting AI chatbots search, select, and pay for products without a human clicking "confirm." In a report released Tuesday, the banks laid out principles for how the technology should develop, and warned that right now, it's advancing faster than the guardrails meant to protect the people using it.

This isn't a hypothetical trend. Technology companies including OpenAI, Anthropic, Google and Meta are actively pitching AI chatbots as shopping tools, building toward a future where an agent picks the product and completes the checkout on your behalf. Retailers have noticed. They're now racing to influence how these chatbots make recommendations, essentially trying to get their products noticed by an algorithm instead of a human browsing a store shelf.

The numbers suggest this shift is happening quickly. British retailer John Lewis said in September that searches originating from AI agents had climbed to 2.5% of its traffic, up from just 0.3% a year earlier. That's roughly an eightfold increase in twelve months. Small as those percentages still look, the trajectory is the story here, not the current size.

Where the Risks Hide

Think of an AI shopping agent like a personal assistant you've just hired, one who has access to your wallet but whose judgment you haven't fully tested yet. The banks' report found that customers are genuinely enthusiastic about this idea. People like the convenience. They want to enable it. But enthusiasm and trust are not the same thing, and the report captures that tension directly.

"Consumers are unclear if AI will act in their interests," the report said. That uncertainty isn't abstract anxiety. It reflects a real gap in how these systems are built and governed right now.

The specific risks the banks flagged are worth sitting with. AI agents may request a customer's card details and enter them directly into websites, a process that removes the usual pause where a human double-checks a suspicious page or an unfamiliar merchant. Agents might also steer users toward payment methods that offer weaker fraud protections, perhaps because those methods are cheaper or faster to process, not because they're safer for the shopper. Layer on top of that the plain worry that an agent might simply buy the wrong item, or spend more than intended, and you start to see why banks are treating this as a consumer protection issue rather than just a technology story.

"They are concerned that AI agents may buy the wrong thing or spend too much, or even worse, lose their money to scams and fraud," the report continued. "They are not sure whether they will be protected or who they will need to go to if things go wrong."

That last line matters most. In traditional fraud cases, there's usually a clear path: call your bank, dispute the charge, file a report. When an AI agent makes a bad decision or gets tricked by a scam site, who's responsible? The retailer? The AI developer? The bank that processed the payment? Right now, nobody has a clean answer, and that ambiguity is exactly the kind of gap that scammers tend to exploit.

It's a pattern that echoes other corners of the AI boom. A separate survey found that three out of four Americans believe AI companies aren't doing enough to prevent serious harm from their products, a sentiment that tracks closely with what banks are describing here. Trust in the technology is running ahead of trust in the systems meant to keep people safe while using it.

Why It Matters

The banks aren't just raising concerns publicly. They plan to bring a set of concrete proposals to policymakers, and the specifics give a sense of how seriously they're treating this. They want disclosure requirements, so consumers know clearly when an AI agent, rather than a human, is handling a transaction. They're pushing for greater transparency around how these agents actually make decisions, since right now the reasoning behind a chatbot's product pick or payment choice is often a black box even to the shopper it's supposedly serving. And they want firmer safeguards around customer data, given how much personal and financial information an agent needs to complete a purchase.

The report also calls for something that sounds almost old-fashioned in the AI era: choice. Consumers and merchants should be free to pick which AI-powered shopping services they use, rather than getting locked into whatever system a retailer or platform decides for them. Different AI systems should also be able to work together, or as the report puts it, be interoperable, so that switching tools doesn't mean starting from scratch or losing protections along the way.

Here's why this deserves attention beyond the finance pages. Agentic commerce is being built on a foundation of convenience and speed, two things consumers genuinely want. But financial protection has historically been built on friction: the pause before a payment, the human verification, the chance to catch a mistake before it becomes a loss. Strip out that friction without replacing it with something equally reliable, and you get a system that's faster but also more fragile.

The banks pushing this report aren't anti-innovation. They're flagging a timing problem. The technology is moving into people's wallets before the rules of the road are settled, and once that horse leaves the barn, retrofitting protections tends to be a lot harder than building them in from the start. For a technology just starting to touch how people pay for groceries, clothes and gifts, getting the basics right early isn't optional. It's the whole point.