Key takeaway: No shared rule yet decides who pays when an AI shopping agent buys the wrong item. Issuers and payment firms are setting their own terms. Merchants may see disputes over agent-initiated orders, and some may see their agent traffic blocked.
Payment companies are taking different approaches to who bears the loss when an AI shopping agent buys the wrong thing. At least one large bank has warned customers they could be liable. Payments Dive first reported these findings on Oct. 28, and CX Dive republished the story on Nov. 3. The dispute centers on agentic payments, where an AI agent initiates a purchase instead of the cardholder.
American Express requires a cardholder to register their agent with the company. If a purchase goes wrong, the cardholder must submit a “purchase intent” document. That document has to show the transaction deviated from what the cardholder intended.
Stripe’s consumer purchase protection covers US purchases made through its Link digital wallet. It carries significant caveats, including monitoring of the agent’s activity. Wells Fargo sent customers an email on Oct. 2 warning they could be liable for an AI agent’s mistakes. That liability applies if they share their account credentials with the agent.
Amazon has blocked Meta’s Muse shopping agent from its retail site. Several banks have also blocked agent traffic, according to the report. Meta released Muse in the US and Canada in September. The company said shopping is the top task among Muse’s 10 million early users.
Early Warning Services CEO Cameron Fowler spoke at the Money20/20 conference. He said the industry needs liability rules that cover agentic commerce broadly, instead of a handful of companies setting their own terms. PNC CEO Bill Demchak said tech companies would need to accept a “massive liability shift.” Stripe card issuing product lead John Piazza said there is no case law yet on agent liability.
Why it matters: Existing dispute processes generally assume a cardholder who either made a purchase or did not. An AI shopping agent that buys the wrong item blurs that line. Each issuer’s terms differ on who has to prove intent, which raises new chargeback liability questions for processors. Merchants that accept agent-initiated orders may need clear records of how they authorized the agent. The debate also touches on AI agent fraud and agentic commerce fraud, since issuers have not agreed on how to tell an agent’s mistake from a bad actor abusing one.
Sources: Payments Dive, and CX Dive