Artificial intelligence is moving online shopping from a system in which consumers make every important decision to one in which software can increasingly make those decisions for them. AI shopping agents can search for products, compare prices, interpret preferences and, in some cases, complete purchases without requiring the customer to manually navigate websites. That shift promises greater convenience, but it also changes where responsibility and risk sit within the transaction.

The concern raised by a group of major banks is therefore less about whether AI can improve online shopping and more about what happens when an autonomous system is given access to money, personal information and purchasing authority. Banks including NatWest, Bank of America, ING, Capital One, ASB Bank and Commonwealth Bank of Australia have warned that agentic commerce is developing faster than the safeguards designed to protect consumers. Their concerns include scams, fraudulent transactions, inappropriate purchases, weak payment protections and misuse of sensitive data.

The warning comes as technology companies and retailers increasingly prepare for a shopping environment in which AI becomes an intermediary between consumers and merchants. Evidence of that transition is already emerging. One major British retailer reported that searches originating from AI agents had increased substantially within a year, indicating that automated systems are beginning to influence how consumers discover products.

AI Changes the Traditional Fraud Model

Traditional online fraud generally involves a human consumer interacting directly with a merchant. The customer sees a product, evaluates the seller, enters payment information and confirms the purchase. Although fraud can occur at any stage, the consumer remains the central decision-maker and can often recognize warning signs before completing a transaction.

AI agents complicate that model because they introduce another decision-making layer. An agent may interpret a user’s instructions, search multiple websites, assess prices and product descriptions, and select an option based on information that the consumer never directly reviews. The system can potentially act faster and across more websites than a human shopper, but speed does not necessarily mean that its judgement is reliable.

This creates a new target for fraudsters. A scammer does not necessarily need to persuade the consumer directly if an AI system can be manipulated into recommending a fraudulent seller. Fake retailers, misleading product information, artificially attractive prices and manipulated online content could potentially influence automated shopping systems. Payment companies have already warned that fraudsters are adapting their methods to the emergence of agent-based commerce, including attempts to make fraudulent merchants appear legitimate to automated systems.

The problem is particularly difficult because an AI agent may successfully complete a transaction while still failing to achieve the consumer’s actual intention. A shopper may instruct an agent to find the cheapest suitable product, for example, but the system may interpret suitability differently from the consumer. The result could be an unwanted purchase rather than a conventional unauthorized transaction.

Access to Financial Data Raises the Stakes

The second major risk comes from the amount of information an effective shopping agent may require. To act on behalf of a consumer, an AI system could need access to purchasing preferences, delivery information, account details, payment credentials and records of previous transactions. The more capable the agent becomes, the greater the potential value of the information surrounding it.

This creates a fundamentally different privacy challenge from ordinary online shopping. A conventional retailer may know what a customer purchased from that particular platform. An AI shopping assistant could potentially build a much broader picture of the consumer by combining information from several services. Shopping habits, financial preferences, household requirements and personal interests could become part of the same decision-making system.

The risk is not limited to deliberate misuse. Data can be exposed through poor security, excessive permissions, compromised accounts or weaknesses in the connections between AI agents and online services. Recent disputes involving autonomous AI tools and major online platforms have demonstrated how difficult it can be to establish acceptable boundaries when third-party agents interact with commercial websites.

Payment information creates another vulnerability. Banks have specifically raised concerns about AI agents requesting or entering card details directly on websites and potentially directing consumers towards payment methods that provide weaker protections. A system designed to optimize convenience could therefore unintentionally prioritize a transaction route that leaves the customer with fewer options if something goes wrong.

The Responsibility Gap Is Becoming More Complicated

One of the most difficult questions surrounding agentic commerce is not simply whether an AI agent can make an error, but who should be responsible when it does. A conventional online purchase normally involves identifiable parties: the consumer, the merchant and the payment provider. Introducing an AI agent adds another participant whose decisions may be generated through software rather than direct human instruction. If the agent purchases the wrong product, selects a fraudulent merchant or exceeds the consumer’s intended spending limit, responsibility can become difficult to establish.

This issue is becoming increasingly relevant as payment companies develop systems specifically designed to allow AI agents to transact. New approaches include virtual payment credentials, spending limits and transaction permissions intended to ensure that an AI system cannot freely access a consumer’s entire financial account. Such controls could reduce exposure, but they also demonstrate that agentic commerce requires a new layer of authorization between the consumer and the payment system.

Consumer protection rules are also having to catch up. Regulators have increasingly recognized that agentic AI can act autonomously across multiple services, creating risks that do not exist when AI merely provides recommendations. Existing consumer law may still apply to transactions involving AI, but applying those principles becomes more complicated when the consumer, AI provider, merchant and payment company each control different parts of the transaction.

Transparency Will Determine Whether Agentic Shopping Scales

The banks’ proposals point towards a broader issue: consumers need to know when AI is acting on their behalf and what authority it has been given. Disclosure alone, however, may not be enough. Consumers also need meaningful information about why an agent selected a particular product, seller or payment method and what data was used to reach that decision.

This is particularly important because retailers have a commercial incentive to influence AI recommendations. In conventional digital commerce, companies compete for visibility through advertising, search rankings and promotional placement. As AI becomes the interface through which consumers discover products, merchants will increasingly seek influence over the systems that make recommendations.

That could create a new form of commercial competition in which the consumer sees fewer products directly and relies instead on an AI-generated shortlist. If the agent’s recommendations are influenced by commercial relationships that are not clearly disclosed, consumers may find it difficult to distinguish between a recommendation based on their interests and one shaped by the interests of a merchant or platform.

The development of agentic commerce therefore requires safeguards that extend beyond payment security. Strong authentication, restricted permissions, transparent recommendations, clear disclosure of AI involvement and effective mechanisms for disputing transactions will become increasingly important. Interoperability also matters because consumers should not become permanently dependent on one AI provider simply because that system controls access to particular merchants or payment networks.

AI shopping agents could eventually make online commerce considerably easier, particularly for routine purchases and price comparisons. But their success will depend on whether convenience can be combined with sufficient control. Giving software the authority to shop is fundamentally different from asking software for a recommendation because the consequences of an incorrect decision can extend directly into a consumer’s finances and personal data.

The banks’ warning highlights that distinction. The technology is developing rapidly, but trust in autonomous commerce will depend on whether financial institutions, technology companies, retailers and regulators establish clear boundaries before consumers begin treating AI agents as trusted purchasing representatives rather than simple digital assistants.

(Adapted from EuroNext.com)

Categories: Economy & Finance, Strategy