Banks warn AI shopping bots raise scam, fraud and data-privacy risks

By Elizabeth Howcroft

PARIS, Sept 22 (Reuters) – Using AI agents for online shopping could increase the risk of scams, fraud and data-privacy breaches, banks including NatWest and Bank of America said ​on Tuesday, as they set out principles for developing ‌the technology.

Technology companies including OpenAI, Anthropic, Google and Meta are increasingly promoting AI chatbots as shopping tools, envisioning a future in which shoppers use AI agents to select products and make purchases on their behalf. Retailers, meanwhile, are ‌racing ​to influence chatbots’ recommendations.

British retailer John ⁠Lewis said in September that searches ⁠originating from AI agents had risen to 2.5% from 0.3% a year earlier, with the trend accelerating.

The group of banks, which also includes ING, New Zealand’s ASB Bank, US lender Capital ​One and Commonwealth Bank of Australia, said in a report that customers were enthusiastic about the potential of agentic commerce and ⁠keen to enable it.

However, they warned ⁠that the technology was advancing faster than industry ​standards and consumer protections.

“Consumers are unclear if AI will act in ​their interests,” the report said.

“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. They are not sure whether they will be protected or who they will need to ⁠go to if things go wrong.”

The report highlighted risks including AI agents requesting customers’ card details and entering them directly into websites, or ⁠steering users towards payment ‌methods that offer weaker protections.

The banks plan ⁠to discuss a series of proposals with policymakers, ​including ‌requiring disclosure when an AI agent is involved ​in a ⁠transaction, greater transparency over how AI agents make decisions, and safeguards to protect customer data.

Consumers and merchants should also be free to choose which AI-powered e-commerce services they use, while different systems should be interoperable, the report said.

(Reporting by Elizabeth Howcroft in Paris. Editing ​by Mark Potter)