A bank may see what an artificial intelligence shopping agent spent without seeing what the shopper authorized.
At The Clearing House’s annual conference Wednesday (Nov. 18), panelists considered who should bear the loss when an agent exceeds that authority and how existing consumer protection laws apply.
Duncan Douglass, partner and co-lead of Alston & Bird’s Payments and Fintech Group, moderated the discussion with Eli Alper, managing chief counsel for regulatory affairs legal at PNC; Kelvin Chen, senior executive vice president and head of policy at the Consumer Bankers Association; Chris Filburn, senior enforcement counsel at the Office of the New York State Attorney General; Jelena McWilliams, president of corporate, legal and external affairs at Plaid; and Carla Sanchez-Adams, senior attorney at the National Consumer Law Center.
Filburn distinguished a purchase the shopper explicitly approves from one the software makes outside its instructions. A shopper who confirms a proposed transaction bears responsibility if they later regret it. When software exceeds the authority it was given, he would put the loss elsewhere.
“From a legal and policy perspective, liability should lie with whoever is best equipped to prevent the harm or loss in the first place,” Filburn said during the discussion. In this case, it’s the AI provider.
Under existing law, a bank could end up absorbing the loss in some cases, giving it a reason to restrict access, Filburn said. He also questioned whether providers could rely on terms disclaiming responsibility.
“And now we’re looking for somebody to be responsible,” Douglass said.
McWilliams drew on the way families already delegate spending. A card issuer knows who is an authorized user, even if it doesn’t know what the cardholder told that person they could buy. Banks should permit agent access through trusted institutions and authorized interfaces, provided they have a reliable way to know what the shopper permitted. Banks could also use existing fraud systems to flag an unusual purchase.
“We shouldn’t underestimate the power of data and fraud detection systems to flag transactions that deviate from normal patterns, place a temporary hold, and ask for verification,” McWilliams said.
Sanchez-Adams pointed to how much broader the permission can be when an agent gets a bank login. With account credentials, a tool could see transaction history or gain access to linked accounts and additional ways to move money. Alper also questioned what consumers infer from the word agent.
“It’s not clear to me if this is an obligation to act in your best interest as opposed to the software’s best interest, or what it can do with the data,” Alper said, adding he favors secure interfaces that let customers control permissions and bind providers to specific limits.
The Payment Method Shapes the Dispute
Sanchez-Adams pointed to the Fair Credit Billing Act’s billing error process. A credit card charge for goods the consumer or their designee didn’t accept, or that weren’t delivered as agreed, can qualify as a billing error. Regulation E covers unauthorized or incorrect electronic fund transfers, including covered debit card and ACH transactions. Its error categories concern the transfer itself. Sanchez-Adams said she sees room for the Consumer Financial Protection Bureau to clarify when an agent’s departure from instructions falls within these protections.
Chen pointed to the official interpretation of Regulation E. When consumers give someone an access device and authority to make transfers, they generally bear responsibility if that person exceeds the authority granted, until they tell the institution that the person no longer has permission. Chen offered that rule as an analogy for delegated payments.
“We are not in a position to understand the nature of the conversation between the user and the agent, or the underlying relationship,” Chen said.
Reliability gives the liability discussion some urgency. Douglass cited Visa’s November shopping benchmark, in which AI models failed 42.8% of simulated runs. The test required models to determine what a shopper needed and refrain from buying if nothing met the requirements. Filburn said differences in performance between models could figure in disputes if providers give shoppers little information about the differences.
Sanchez-Adams said arbitration clauses and waivers in AI providers’ terms could limit consumers’ ability to sue. Chen pointed to Dodd-Frank as a possible basis for public enforcement when a company takes unreasonable advantage of consumers who reasonably rely on a covered person to act in their interests. McWilliams said she expects state attorneys general and courts to shape how these disputes are handled before new federal rules are finalized.
For all PYMNTS AI coverage, subscribe to the daily AI newsletter.