AI shopping is moving from "help me choose" to "buy it for me."
That sounds like a small change, but it is actually a major shift. A chatbot that recommends a pair of headphones is one thing. An AI agent that compares stores, enters checkout, selects a payment method and completes the purchase is acting with real money and real consequences.
That is why a new argument is opening up around agentic commerce. Technology companies and payment networks see AI shopping agents as the next layer of online retail. Banks and consumer advocates are asking a more uncomfortable question: who is responsible when the agent gets it wrong?
The answer matters because the technology is already moving into real shopping systems.
What is an AI shopping agent?
An AI shopping agent is software that can do more than answer questions about products.
Depending on the system and the permissions you give it, an agent may be able to:
- search several stores for an item;
- compare prices, delivery times and product specifications;
- remember preferences such as size, brand or budget;
- place products in a cart;
- choose a payment method;
- complete a purchase on your behalf.
The important word is permission.
Traditional e-commerce assumes a human is present for most important steps. You search, click, review the cart and confirm payment. Agentic commerce tries to automate more of that chain.
That can save time. It also creates new points where a misunderstanding, attack or bad decision could become a real transaction.
Why AI shopping is suddenly becoming a big deal
The infrastructure is catching up with the idea.
Google has been building what it describes as the foundation for agentic commerce, including its Universal Commerce Protocol and Universal Cart. The goal is to let AI-powered shopping experiences search, coordinate and eventually transact across a much larger part of the online retail ecosystem.
Mastercard is moving in the same direction. It has been developing tools such as Agent Pay and Verifiable Intent, which are designed to let AI agents make authorized transactions while preserving a record of what the user actually approved.
This is not only a technology-company experiment. According to Reuters, British retailer John Lewis said AI-originated searches increased from 0.3% to 2.5% over a year.
That is still a small share of shopping traffic, but the direction is clear: more consumers are beginning product discovery inside AI systems rather than traditional search boxes.
Why banks are worried
On September 22, Reuters reported that major banks including NatWest, Bank of America, ING, Capital One, Commonwealth Bank of Australia and ASB were warning about the risks of AI-powered shopping.
Their concern is not simply that AI might recommend the wrong product.
The deeper problem is that an autonomous shopping agent can sit between the customer, the merchant, the bank and the payment network. If something fails, responsibility can become less obvious.
Banks highlighted concerns around fraud, scams, privacy, financial-data handling and consumer protection.
Those worries can be grouped into four practical risk areas.
Risk 1: the agent misunderstands what you authorized
Imagine telling an agent:
Find the cheapest direct flight under $600 and book it if baggage is included.
A human can notice that the only available option has a long overnight connection or a restrictive refund policy. An automated agent might interpret "direct" or "baggage included" differently than you expected.
For a low-cost household item, that may be an annoyance.
For travel, electronics, subscriptions or expensive purchases, a small misunderstanding can have a larger financial effect.
This is why payment companies are working on systems that preserve verifiable intent: a record of exactly what a consumer allowed the agent to do.
Risk 2: AI agents create a new target for scams
AI agents could become valuable targets for attackers because they may have access to shopping preferences, account sessions, payment permissions or saved credentials.
The danger is not necessarily that the AI itself is malicious.
An attacker might instead try to manipulate what the agent sees.
For example, a fake store could be designed to look attractive to automated shopping systems. A compromised page might feed misleading product information to an agent. A phishing message could try to persuade a user to grant broader purchasing permissions than intended.
This creates a new version of an old security problem: criminals follow the money, and shopping agents may become another path to it.
Risk 3: privacy becomes more complicated
A good personal shopping agent becomes more useful when it knows more about you.
That can include:
- your preferred brands;
- clothing sizes;
- delivery address;
- purchase history;
- travel plans;
- price limits;
- family preferences;
- payment options.
Individually, those details may look harmless. Together they create a detailed consumer profile.
The privacy question is not only "is my card number encrypted?"
It is also:
Who gets to see the data used to make the decision?
If an AI agent checks several merchants, comparison services and payment systems, information may move across multiple companies during one task.
Consumers may not always know which parts of their request are shared, stored or used for personalization.
Risk 4: refunds and liability may become confusing
Online shopping already produces disputes over refunds, chargebacks, subscriptions and unauthorized transactions.
Now add an AI agent.
Suppose the agent purchases an item that technically fits the instructions but is not what the customer expected.
Was the transaction authorized?
Should the merchant refund it?
Is the AI provider responsible?
Does the bank treat it like a normal card transaction?
What if the agent was manipulated by a fraudulent seller?
These are not purely technical questions. They affect payment rules, consumer rights and customer support.
Reuters reported that banks want clearer disclosure when AI agents are involved in transactions and stronger protections around responsibility and consumer recourse.
The Amazon-Meta dispute shows the trust problem is already real
One of the most useful examples is the conflict between Amazon and Meta's Muse shopping agent.
Current reporting says Amazon blocked Muse from making purchases on its platform, with the dispute involving unauthorized agent activity, privacy, identification and platform-control concerns.
Whatever side you take, the episode exposes an important problem.
For agentic commerce to work smoothly, websites need to know who or what is accessing them, consumers need to know what an agent is doing, and payment providers need to know whether the transaction reflects genuine user intent.
The technology can automate a shopping journey, but trust cannot simply be assumed.
What companies are building to make agentic shopping safer
The industry is not ignoring these risks.
Several safeguards are emerging.
Spending limits
A consumer could allow an agent to spend up to a fixed amount rather than giving unlimited purchasing authority.
That makes a $40 mistake very different from a $4,000 mistake.
Restricted merchants or product categories
Agents can be limited to approved stores, brands or types of products.
A business procurement agent, for example, might be allowed to order only from a pre-approved supplier list.
Tokenized payment credentials
Instead of giving an AI agent raw card details, payment systems can provide controlled credentials or tokens designed for a specific transaction or purpose.
That reduces the damage if a credential is exposed.
Verifiable intent
Mastercard has described Verifiable Intent as a way to create an auditable connection between the user's permission and the action an AI agent takes.
The idea is simple: if the agent buys something, the system should be able to show what the customer actually authorized.
Agent identification
Merchants may increasingly need to distinguish between human shoppers, legitimate AI agents and abusive automated systems.
That could become as important to e-commerce as identifying trusted payment processors is today.
Are AI shopping agents safe today?
There is no useful one-word answer.
An AI agent that searches for products and recommends options carries relatively little financial risk.
An agent that can autonomously spend money deserves much stricter controls.
The safety of the system depends on several layers working together:
- how much permission the user gives the agent;
- how payment credentials are protected;
- whether merchants can identify legitimate agents;
- whether the user's intent is recorded clearly;
- how disputes and refunds are handled;
- how much sensitive information moves between services.
The important distinction is between AI-assisted shopping and fully autonomous purchasing.
The second requires a much higher level of trust.
A practical checklist before letting an AI buy for you
If you use an AI shopping agent, treat purchasing permission the same way you would treat access to a financial account.
Start with a low spending limit
Do not give a new agent broad authority on day one.
Test it with inexpensive purchases first.
Require confirmation for expensive transactions
Automation is useful, but there is little benefit in removing the final confirmation step for a major purchase.
Prefer tokenized or virtual payment methods
A payment credential with limited scope is safer than giving a service unrestricted access to a primary card.
Review what data the agent stores
Check whether the service keeps purchase history, addresses, messages or payment-related information.
Use reputable merchants
An AI system can compare thousands of results quickly, but speed does not automatically mean that every seller is trustworthy.
Keep transaction notifications enabled
Real-time payment alerts are one of the simplest ways to catch an unexpected purchase quickly.
The bigger question: who will control the shopping relationship?
Security is only part of the debate.
AI agents could also change the balance of power in online retail.
Today, a marketplace controls much of the shopping experience: search results, ads, recommendations, reviews and checkout.
If consumers begin by telling an independent AI agent what they want, the agent may decide which stores and products the user ever sees.
That gives AI platforms enormous influence.
Amazon's decision to block an outside shopping agent and Shopify's willingness to work with agentic systems illustrate two very different strategies for the same future.
One protects the marketplace boundary. The other tries to become infrastructure inside the agent-driven shopping journey.
For shoppers, the real question may eventually become:
Is my AI agent working for me, for the retailer, or for whoever pays to influence its recommendations?
That question could matter as much as fraud protection.
Bottom line
AI shopping agents could remove a huge amount of friction from online shopping.
They can compare products faster than a person, monitor prices, follow detailed preferences and potentially handle routine purchases automatically.
But once an AI system can spend money, convenience is no longer the only metric that matters.
Authorization, privacy, security, transparency and responsibility become part of the product.
Banks are right to ask how consumers will be protected when something goes wrong. Technology and payment companies are also right that many of those risks can be reduced with better identity, tokenized payments, spending controls and verifiable permissions.
The winners in agentic commerce will probably not be the systems that buy the fastest.
They will be the systems consumers trust enough to let them buy at all.
Sources
- Reuters — Banks warn AI shopping bots raise scam, fraud and data-privacy risks
- Google — Introducing the Universal Cart and more ways to help you shop
https://blog.google/products-and-platforms/products/shopping/google-shopping-cart/
- Mastercard — Mastercard’s vision for trusted agentic commerce
- Mastercard — How Verifiable Intent builds trust in agentic AI commerce
https://www.mastercard.com/us/en/news-and-trends/stories/2026/verifiable-intent.html
- The Verge — Amazon blocks Meta’s Muse AI agent
https://www.theverge.com/tech/998078/amazon-blocks-meta-muse-ai-agent-shopping
