Banks and payments companies are being forced to work out questions of errors and liability as they prepare for agentic commerce.
The generative artificial intelligence that powers agentic commerce — including the AI agents that search for products online and may eventually make purchases on consumers’ behalf — is prone to mistakes, and questions remain about how to resolve that problem.
“Could the agent hallucinate and buy something we didn’t tell it to buy?” said Mike Lozanoff, global head of merchant services for JPMorgan Chase, the biggest U.S. bank.
If that happens, it’s not clear who is responsible for fixing it, he said during an interview Monday on the sidelines of the Money 20/20 conference in Las Vegas. “The rules here are not fully formed yet,” Lozanoff said.
The issue was a frequent topic of discussion at panels and in interviews with payments executives at the conference.
Payments companies are still working out how to handle these issues when they inevitably arise, said Serge Elkiner, general manager for Paze, a peer-to-peer payment platform owned and operated by bank-owned Early Warning Services.
When a shopper “authorized that agent to do a transaction, who says that agent bought the right thing?” Elkiner said in an interview at the conference.
If a consumer tells an AI agent to order a navy blue T-shirt and gets a red one instead, they might call the bank, which doesn’t have a record of that specific consumer ordering the shirt because the agent made the purchase, he said.
“So then they call the shop, but the shop doesn’t have a record of it either,” Elkiner said. “All of these things are operational headaches.”
Early Warning Services, which also owns and operates Zelle, and other payment companies are trying to build the infrastructure to address those problems and hopefully alleviate those headaches, he said.
Agentic AI is mostly used to search for products under certain conditions — think a specific product that a consumer needs in a certain timeframe and above or below specific price points — but will eventually be empowered to make transactions on a shopper’s behalf.
When that time arrives is an open question, but it’s an eventuality that payments players must prepare for, Elkiner said.
Consumers are already using generative AI programs like OpenAI’s ChatGPT and Perplexity to find products, noted Michelle Gill, an executive vice president at PayPal Holdings, who is also general manager of the small business and financial services group. AI that does your shopping for you isn’t far off, she said.
PayPal is moving to weave itself deeply into the emerging agentic AI ecosystem, announcing partnerships Tuesday with OpenAI to connect its global merchants network with ChatGPT for sales and to embed its digital wallet to enable transactions within the AI provider’s signature tool.
PayPal said separately that it will launch a suite of new commerce tools to help merchants address agentic commerce, including a payment solution, a catalog and order-management tool to help retailers “connect product data, inventory, and fulfillment with AI-driven discovery and checkout experiences.”
“Arguably an agent, appropriately trained over time, will make less mistakes than a human,” Gill said in an interview. But in the meantime, “the infrastructure and architecture has been set up for human mistakes.”
Countering that problem means building new payments architecture and infrastructure, she said.
