AI helped Uncle Sam catch $1 billion of fraud in one year. And it’s just getting started

The federal government’s bet on using artificial intelligence to fight financial crime appears to be paying off.

Machine learning AI helped the US Treasury Department to sift through massive amounts of data and recover $1 billion worth of check fraud in fiscal 2024 alone, according to new estimates shared first with CNN. That’s nearly triple what the Treasury recovered in the prior fiscal year.

“It’s really been transformative,” Renata Miskell, a top Treasury official, told CNN in a phone interview.

“Leveraging data has upped our game in fraud detection and prevention,” Miskell said.

The Treasury Department credited AI with helping officials prevent and recover more than $4 billion worth of fraud overall in fiscal 2024, a six-fold spike from the year before.

US officials quietly started using AI to detect financial crime in late 2022, taking a page out of what many banks and credit card companies already do to stop bad guys.

The goal is to protect taxpayer money against fraud, which spiked during the Covid-19 pandemic as the federal government scrambled to disburse emergency aid to consumers and businesses.

To be sure, Treasury is not using generative AI, the kind that has captivated users of OpenAI’s ChatGPT and Google’s Gemini by generating images, crafting song lyrics and answering complex questions (even though it still sometimes struggles with simple queries).

Instead, the fraud detection efforts rely on machine learning, the subset of AI that excels at analyzing vast amounts of data, and making decisions and predictions based on what it’s learned.

AI can be very helpful in fighting financial crime by combing can through almost endless streams of data and detecting subtle patterns – all in a fraction of the time it would take a human to do it. Experts say that once sophisticated AI models are trained, they can sniff out suspicious transactions in mere milliseconds.

“Fraudsters are really good at hiding. They’re trying to secretly game the system,” Miskell said. “AI and leveraging data helps us find those hidden patterns and anomalies and work to prevent them.”

This is especially crucial for the Treasury, which is among the biggest payers on the planet – if not the biggest.

Each year, the Treasury delivers about 1.4 billion payments valued at nearly $7 trillion to 100 million people. It’s responsible for sending out everything from Social Security and Medicaid payments to federal worker paychecks, tax refunds and stimulus checks.

That critical role makes Treasury a prime target for fraudsters seeking to steal from taxpayers.

Last year, the Internal Revenue Service announced it has deployed AI to detect tax cheats by examining large and complex returns from hedge funds, law firms and others.

Online payment fraud is expected to surpass $362 billion by 2028, according to estimates from Juniper Research.

And some of that fraud is being turbocharged by AI itself.

In one infamous case earlier this year, Hong Kong police say a finance worker was tricked by a deepfake video into paying $25 million to fraudsters.

US officials have expressed concern that AI introduces new dangers into the financial system. Treasury Secretary Janet Yellen in June warned bankers that AI in finance poses “significant risks.”

Top regulators, led by Yellen, classified AI late last year as an “emerging vulnerability” to the financial system.

Miskell stressed that while AI systems will flag suspicious transactions, a “human is always in the loop” and federal agencies make the final determination of whether something constitutes fraud.

The Treasury’s use of AI to fight financial crime is just getting started.

Miskell indicated the Treasury is exploring how to adopt the fraud-detection methods that leading banks and credit card companies are deploying, declining to go into detail to avoid “tipping off bad actors.”

A Treasury spokesperson told CNN that the department is speeding up its work to enhance the fraud-detection tools available to federal- and state-administered programs. Officials are testing new data sources to better spot fraud and shady payments, and they are teaming up with state agencies to fight unemployment insurance fraud.