News analysis ·

Plaid's Perret on AI cards: who earns and who pays for fraud

Plaid's CEO expects AI platforms to issue cards and warns of a scary fraud year. We trace who earns the interchange and who absorbs the losses.

Close-up of a card payment terminal screen showing an amount, a contactless tap symbol and a prompt to insert, swipe or try another card
Photo: User:Daylon124 / Wikimedia Commons, CC0 (cropped)

Plaid chief executive Zach Perret told the Money20/20 conference in Las Vegas that AI platforms will likely start making money from financial services, beginning with cards issued inside their apps, Banking Dive reported. In the same talk he said he expects a scary year for security. This article looks at how money would flow through an AI-issued card, and who would carry the losses when something goes wrong.

What Perret said at Money20/20

Plaid is a financial data aggregator based in San Francisco. It links customers' bank accounts to other apps and works with about 12,000 institutions, according to Banking Dive. That gives its chief executive a wide view of where account data, and money, is heading.

Perret said AI platforms, naming ChatGPT from OpenAI and Claude from Anthropic, are likely to treat finance as a revenue source. He calls these platforms harnesses. His first example was card issuance: a card that lives inside the AI platform, with the platform collecting the interchange fees. Lending products could come later, he said, but cards are the likely starting point.

He did not expect one firm to own the whole customer. Simple jobs, such as moving money between checking and savings, might shift to AI platforms. A mortgage is something customers may still want to handle directly with the bank. He also said a neobank that issues mortgages may want to keep that conversation on its own platform rather than inside an AI tool.

How the money moves on an AI-issued card

Interchange is the fee a merchant's bank pays to the bank that issued the customer's card on each purchase. It is the main reason card issuing is a business at all. When a platform says it will collect interchange, it usually means it is sharing that fee with a licensed bank that actually issues the card behind the scenes.

The chain would look roughly like this. The customer asks the AI platform to pay. The platform uses an embedded card. The merchant's bank, called the acquirer, sends the transaction over the card network to the issuing bank. The issuing bank approves it, pays the merchant side, and receives interchange. The platform then gets its agreed share of that fee from the issuer. The customer repays the issuer, either from a linked account or on a credit line.

So the money still settles through banks and card networks. What changes is who owns the customer screen, and who earns a slice of the fee for owning it.

Who is regulated, and who bears the losses

In this setup, the regulated party is the bank that issues the card. It holds the account, follows card network rules and answers to bank supervisors. The AI platform is a front end. Its duties come mostly from its contract with the bank, not from a banking licence.

That split matters for fraud. When a card payment is disputed, the issuer is usually the first to face the claim, and then sorts out costs with the merchant side through network rules. Depending on the contract, the issuer may try to pass some of that cost back to the platform. A customer who did not authorise a payment would normally deal with the issuing bank, even if they only ever saw the AI app.

Perret was blunt about the threat. Deloitte has projected that U.S. fraud losses could reach $40 billion next year because of generative AI. Perret called that number a conservative estimate. He said swarms of AI agents attacking banks worry him more than deepfakes. "We're just in a dead sprint using AI to fight AI fraud," he said. Plaid builds fraud tools on its own data network and works with partners, and he described the fight as a joint industry effort that will last a few years.

Where banks fit in

Banks are not standing still. Fifth Third chief executive Tim Spence said incumbents will build services that customers can reach through Meta's Muse, with bank-grade information security, Banking Dive reported. Perret also argued that banks should let customers chat with their finances inside the bank app and bring in outside data there.

Fitch Ratings sees AI disruption as a moderate risk to banks over the coming five years, mostly through tech competition on processes and products. It said banks serving households and small and midsize firms are the most exposed, because their standardised, high-volume products are easy to copy. How fast customers adopt AI-led services is the key caveat.

What to watch

Perret's comments are a view, not an announcement. The useful questions are about plumbing and liability, not about which app wins.

  • Whether any AI platform announces an issuing bank partner, and how the interchange split is described.
  • How disputes and unauthorised payments are handled when the customer only dealt with an AI app.
  • Whether banks launch their own chat tools before AI platforms win the everyday tasks.
  • Signs that fraud losses are rising as Perret warned, and who ends up paying for them.

Sources

Written by AI from the sources above and published automatically: before release, a script confirms every figure against the sources and checks for copied text. No editor reviews it. Treat it as background reading, not financial advice.

← All posts