NatWest and Bank of America have warned that AI shopping agents could increase consumers’ exposure to scams, fraud and data-privacy breaches as banks call for clearer safeguards around autonomous purchases.
The two lenders are part of a five-bank group with ING, ASB Bank and Capital One that has set out principles for agentic commerce: systems in which artificial intelligence can move beyond recommending products and take steps towards selecting goods, choosing payment methods and completing transactions for a customer.
The legal difficulty begins with authority. A consumer may tell an AI agent what to buy and set a budget, but disputes become more complex if the agent exceeds those instructions, selects the wrong product or routes the payment in a way the customer did not expect. Lawyers advising banks, retailers and AI providers will need to consider how consent is recorded, what evidence shows the scope of the agent’s mandate and how responsibility is allocated if a transaction is challenged.
The banks have also highlighted fraud and data-protection risks. One concern is that an agent may request card information and enter those details directly into a merchant website. Another is that automated systems could steer customers towards payment methods offering fewer protections. The group is therefore calling for greater transparency over how agents make decisions, clear disclosure when AI is involved in a transaction and stronger safeguards for customer information.
Those questions sit alongside a wider UKJT AI Liability Statement — Lawyer Monthly issue. Existing payment, AliExpress Fined €550m Under EU Digital Services Act and privacy obligations do not disappear because software has been inserted into the transaction. What changes is the chain of actors involved and the evidence needed to determine who authorised a particular step, what data was used and which party should deal with a loss or complaint.
The banks also want consumers and merchants to retain a choice of AI services and are pushing for interoperability between systems. That reflects a competition concern as well as a technical one: if AI agents become a major route into online shopping, control over the agent could influence which products, merchants and payment methods consumers see.
The commercial shift is already visible. John Lewis said in September that the share of product searches reaching its website through AI agents had risen from 0.3% to 2.5% in a year. OpenAI, Google, Anthropic and Meta are among the technology companies developing AI assistants capable of helping consumers search for or select products.
For legal teams, the practical question is increasingly one of auditability. Businesses deploying agentic commerce will need records showing what the customer asked the system to do, what limits were imposed, how the agent reached a decision and when human approval was required. Those records could become important in complaints, chargebacks, fraud investigations and wider disputes over Workday AI Hiring Bias Case Seeks Class Status | LM.
NatWest, Bank of America, ING, ASB Bank and Capital One plan to take their proposals to policymakers. The regulatory challenge will be to preserve the convenience promised by autonomous shopping without allowing uncertainty over consent, data use and liability to become a new source of consumer harm.
As agentic commerce expands, the legal framework may not turn on whether AI is permitted to shop for consumers, but on whether businesses can prove what the system was authorised to do and who remains accountable when it goes wrong.