Plaid launches 'Protect' anti-fraud product as financial fraud rises 25% a year

Summary

Plaid CEO Zach Perret launched Protect at the company's annual customer conference, Plaid Effects. The anti-fraud product analyzes network-wide data across all fintech apps using roughly 20 detection analyses. Perret has been pitching this concept since 2014, when he first argued that Plaid would eventually accumulate enough cross-platform data to build financial analytics no single institution could replicate. The 2022 acquisition of identity verification company Cognito was a key step toward making it real.

Protect runs simultaneous analyses on each transaction or sign-up event. Its clearest value comes from cross-network visibility. If a user signs up for Coinbase today, that is normal. If the same user signs up for six lending products the same day, that is a fraud signal no individual lender would catch alone. Other signals include impossible travel (logging into a fintech app in San Francisco and triggering activity in Singapore two hours later), bank account age, and phone number age. Perret says the hardest part is not the analysis itself but the daily drift in fraud patterns, which requires both algorithmic updates and a manual layer where human analysts search for newly identified fraud rings and write rules against them.

Plaid is not heavily transformer-driven yet, though it is moving that direction. LLMs are used for tooling and data cleansing on the backend rather than core fraud detection. Perret expects every new fraud technique to become table stakes once the industry adopts it, so competitive pressure is constant iteration.

Financial fraud is growing at roughly 25% a year, with AI-driven schemes accounting for a meaningful share of that growth. AI-generated text messages designed to manipulate victims are already part of the mix. Perret argues that personal data should be assumed compromised at this point, and that the only real defense is better tooling. On regulation, he is skeptical that KYC rules requiring banks to retain sensitive data indefinitely will change quickly, but he argues that relying on a static Social Security number as a fraud detection tool is already inadequate regardless of what regulators do.

Plaid faces a more immediate legal concern. The CFPB's rule implementing Section 1033 of Dodd-Frank, which established consumers' right to share their financial data with third parties like Plaid, was finalized under Rohit Chopra and sued by a coalition of banks on the day it was published. The lawsuit appeared prepared in advance. Under the current administration, the OCC has signaled agreement with the banks, and the push is to either pare back or eliminate the rule.

Perret does not object to a rollback itself. His concern is how the dismissal is being structured. The language being used could, in his reading, allow individual banks to selectively cut off data access to industries they find competitive or inconvenient. He draws a direct line to the debanking wave that targeted crypto a few years ago. This outcome would hand banks a legal mechanism to do the same thing to crypto, gaming, or any other sector they choose to target, unilaterally and without coordinated action.

On Circle's IPO and stablecoins broadly, Perret sees current use cases—currency hedging, cross-border transfers, higher-yield savings—as large but not the interesting part. The more significant shift he expects is consumers moving fluidly between bank accounts and crypto wallets, and companies such as payroll processors capturing float that previously sat entirely with banks. That float, earning yield at current interest rates, becomes a new revenue stream, though Perret notes it also introduces rate sensitivity into business models that did not have it before. Large payments platforms have an incentive to offer more tender types including stablecoins to keep more dollars on their own platforms longer.

On whether AI agents will default to crypto payment rails, Perret pushes back on the consensus narrative. Bank accounts already have OAuth. Cards are moving toward digital-native authentication. Crypto is getting there too. His expectation is that within two to three years, agents will have access to all major tender types rather than defaulting to any one of them.