The red coat still bothers Early Warning Services CEO Cameron Fowler.

His son purchased a new red coat, in the wrong size as it turned out, using social media company Meta Platforms’ Muse personal agent, Fowler recalled last week in an interview.

“I don’t know why you want a red coat, but more a red coat that doesn’t fit you and … who’s liable for that?” Fowler said on Oct. 19, recounting the conversation with his son, one of “the early 20-year-olds in my house who are on the front foot” of agentic-powered shopping.

For Fowler, the ill-fitting red coat symbolizes an array of issues confronting the industry in a new era of agentic payments as consumers direct their personal agents to shop for them. The formulation of corporate policies and federal legislation to address topics such as liability, data access and consumer intent has sorely lagged the pace of agent technology, he said on the sidelines of the annual Money20/20 financial technology conference.

“What do you need to have in place to ensure that the least-informed consumer actually understands the commitments they’re making, and that we are very clear on the relationship between human and authorization,” said Fowler, a former BMO Financial Group banker who leads EWS. “That’s where the rubber meets the road.”

With consumers beginning to tap artificial intelligence agents to shop and make payments, the banking industry is increasingly edgy about risks associated with bot traffic inside customers’ accounts. Early Warning Services, parent of the Zelle payment service and Paze digital wallet, is owned by seven large U.S. banks.

Some bank executives are alarmed and incensed by the prospect of machines making payments without liability measures in place, given what they consider the inherent risks. 

The CEO of PNC Financial Services Group, William “Bill” Demchak, railed against tech companies for barreling into agentic transactions without heeding the perils during his Oct. 19 appearance at the Money20/20 conference.

“There’d have to be a massive liability shift that these guys took on the liability associated with breaking all the crap they’re about to break,” he said. “They’re kids in a room with a new toy, and they’re messing around with money … this isn’t a game,” Demchak said. “This isn’t like normal technology where you’re allowed to break stuff and figure it out later. You’re messing with people’s lives. It’s money.” 

Meta’s Muse agent

Fears about the fast pace of agentic commerce were sparked when Meta Platforms released its Muse personal agent in the U.S. and Canada in September. 

Agentic commerce “started going vertical” that same month, Stripe President John Collison said in a podcast discussion earlier this month. Meta said last week that shopping is the top task among Muse’s 10 million early users. Beyond the banks’ agentic blocks, retail giant Amazon has also refused Meta’s agent entry to its shopping sites.

A detailed audit trail for agentic transactions is “extremely crucial” to enable responsible commerce, card network Visa’s head of strategic financial institution and fintech partnerships, Kaitlan Moran, said Oct. 20 at a Money20/20 panel on agentic commerce.

“From our perspective, the really important thing is to make sure that we’re creating that trust layer and providing that enhanced context and data and information to all of the parties in the ecosystem,” Moran said. 

Standards are still under development and “we don’t have case law on liability,” said John Piazza, product lead for digital payment processor Stripe’s card issuing business, and an adjunct professor at Northwestern University’s Pritzker School of Law. Stripe’s Link digital wallet helps to enable payments by the Muse agent.

“There’s still a lot of macro questions,” he said, speaking on the same panel with Moran, about agent-driven payments.

New rules of the payment road 

In the “traditional” four-party payment model of cardholders, merchants, issuers and acquirers, networks influence behavior through interchange rates and chargeback rules, said Andrew Lambert, the head of product for Cross River Bank, a Fort Lee, New Jersey-based banking-as-a-service provider.

A merchant that offered a network more transaction data might benefit from a lower interchange cost or could shift liability on a given transaction by demonstrating tighter security measures, Lambert said, speaking on the same panel as the Stripe and Visa executives.

“When you’re talking about an agent platform that sits outside of that ecosystem, those levers don’t work anymore, and I think that’s the thing that I find the most perplexing,” Lambert said. Card networks should provide “more guidance around rules of the road” for agentic transactions, he added.

Wells Fargo, the fourth-largest U.S. bank by assets, is warning customers that they might be liable for agents’ errors if they share their account details. “Just like humans, AI can make mistakes,” the bank wrote in an Oct. 2 customer email. “If you give an AI tool your username and password, you may be liable for any mistakes made by the AI tool.”

Proliferating protocols

The major card networks, including Visa, Mastercard and American Express, have adopted different protocols for how agents should transact. Visa’s “trusted agent” approach, for example, involves a directory of known agents, while Mastercard’s “agent pay” adopts an authorization process to tokenize payment credentials for agents’ access.

Tech titans Google and Stripe, joined by payment processors, have also introduced agentic protocols to help forge operational standards for these new shopping bots. 

Some of the new infrastructure hasn’t been rolled out yet, which has prompted concerns about liability and other issues, Google’s vice president and general manager of payments, Stavan Parikh, said Monday in an interview.

“What has happened in the ecosystem so far is not leveraging what the networks have fully built out,” he said. “But I expect in the coming weeks and months we will see people moving to more of these standardized infrastructures.”

Two large players in the payments industry — American Express and Stripe – have announced consumer purchase protection policies aimed at agentic transactions. 

Among other terms, Amex’s rules require that the cardholder’s agent be registered with the company. If a purchase goes awry, the purchaser must send a “purchase intent” document to help prove that a particular transaction deviated from the consumer’s intent. Stripe’s plan covers purchases in the U.S. made through Link, with extensive caveats in the user terms about monitoring the agent’s activity.

Still, limited rules from two companies hardly suffice for transactions that could reach into the trillions over time, industrywide, Fowler said.

“I’m sure there are lots of people who are orchestrating wonderful transactions, but here we’re talking about hundreds and hundreds of millions of consumers who have no idea how powerful the thing is and may not have been fully educated on the risks that they’re taking, and that’s not acceptable,” Fowler said.

“At the end of the day, was there a human that authorized (a transaction), and if so, what's the liability? If not, what's the liability? I think that that needs to be in place for the industry, not for a couple of the players.”