For twenty years, ecommerce fraud prevention has asked one question. Is this really the cardholder? Agentic commerce adds a second one. Is this agent really allowed to buy on that person's behalf?
Federal Reserve Governor Christopher Waller put it well in a speech in September 2026, describing the shift as moving toward "proving that an agent has the authority to pay on the buyer's behalf." He also named trust between buyers and sellers as the biggest barrier to agentic commerce. For finance, payments and operations leaders, that is where the practical work sits.
What changes when an AI agent places the order?
Several assumptions behind today's fraud and service processes stop holding.
- No person is in the session. The usual signals, such as how someone types, moves through the site or completes checkout, may not exist.
- Good agents can look like bad bots. Fast, automated, repeated requests are exactly what bot protection is built to block.
- Intent is less clear. If an agent orders the wrong size or the wrong quantity, was that the shopper's instruction or the agent's mistake?
- Mistakes can repeat at scale. A misunderstanding a person makes once, an agent can make on every order until someone notices.
How the card networks are responding
The payment networks are building ways for merchants to tell trusted agents from everything else.
In October 2025, Visa's Trusted Agent Protocol and Mastercard's Agent Pay were announced with Cloudflare, using cryptographic signatures so a merchant can verify that an agent is registered, that it is linked to a real consumer and whether it is browsing or buying. In September 2026, Mastercard expanded Agent Pay with shared trust and intelligence signals and a score, being tested in the US, that indicates how likely it is that a transaction was started by an AI agent.
Google's Agent Payments Protocol takes a related approach. It uses signed records called mandates, one capturing what a shopper asked an agent to do and another capturing the exact cart they approved, so there is evidence of what was authorized. We explain how this fits with the shopping protocols in UCP, ACP and MCP explained.
The bigger risk may be blocking good agents
Most merchants worry about fraudulent agents. The more immediate risk is quieter. Bot protection, firewall rules and fraud filters set up years ago may block or challenge legitimate agents acting for real customers.
Nobody tells you when that happens. The agent simply recommends a competitor whose store answered. Review your bot management and fraud settings with your platform and security providers, and ask how they handle verified AI agents.
Questions to ask your payments provider
- Can you identify agent-initiated transactions? And can we see them separately in reporting?
- Do you support the Visa and Mastercard agent programs? If not, when?
- How do our fraud rules treat agent orders? Are they scored differently, and can we tune that?
- What is our liability position? Who carries the loss when an agent purchase is disputed?
- What changes for chargebacks? What evidence will help us win a dispute involving an agent?
Providers such as Adyen are part of these industry efforts, so these are fair questions to raise at your next review.
Returns, disputes and customer service
When an agent gets an order wrong, the customer will still contact you, not the agent. A few changes make that easier to handle.
- Make policies clear and machine-readable. Agents should be able to read your return window, costs and conditions before they buy.
- Keep the human in the loop. Send order confirmations the person will actually see, and make cancelling within a short window easy.
- Record how the order arrived. Knowing an order came through an AI channel helps customer service resolve disputes faster.
- Watch for patterns. A spike in returns from one channel is often a data problem, such as unclear sizing, rather than fraud.
How fast is this actually happening?
Slower than the headlines suggest, and faster than most fraud rules assume. Shoppers use AI assistants heavily for research, but they have been more cautious about letting agents spend money. OpenAI retired its in-chat Instant Checkout in March 2026 and sent checkout back to merchants' own sites. Mastercard projects that one in ten consumers will routinely use AI agents for shopping and payments by 2030.
That means the near-term effect for most brands is agents sending shoppers to your checkout, not agents paying on their own. The trust work still matters, because the rules and tools are being set now. Business purchasing will likely move first, a point Waller made as well. We look at that in AI agents in B2B ecommerce.
Common questions
Are orders placed by AI agents riskier than normal orders?
Not inherently. A verified agent acting for a known customer can be lower risk than an anonymous checkout. The risk comes from not being able to tell verified agents from malicious bots, which is what the network programs are designed to solve.
Who is liable when an AI agent buys the wrong thing?
It depends on card network rules, your payment provider's terms and your own policies, and it is still being worked out across the industry. Ask your provider directly, and check your terms of sale with your legal advisor.
Should we block AI bots from our store?
Block bad ones, not all of them. Blanket blocking can stop legitimate agents acting for real customers. Look for bot management that can recognize verified agents.
Do we need a new payment integration?
Usually not yet. Much of this is being built into payment providers and commerce platforms. Your job right now is to ask the right questions and make sure your settings are not working against you.
Does this apply to Canadian brands?
Yes. Some programs, such as Mastercard's new agent score, are being tested in the US first, so brands selling into the US have the most reason to prepare now.
Where to start
Put agentic payments on the agenda for your next payments and fraud review, and check whether your bot protection is turning away good agents. If you are working out where AI fits in your ecommerce roadmap, get in touch.
