The federal government’s investment in artificial intelligence for combating financial crime is beginning to show significant results. According to new estimates shared exclusively with CNN, machine learning AI has enabled the US Treasury Department to analyze vast amounts of data, resulting in the recovery of $1 billion in check fraud for fiscal year 2024 alone—nearly three times the amount recovered the previous year. “It’s truly been transformative, ” said Renata Miskell, a senior official at the Treasury, in a phone interview with CNN. “Utilizing data has significantly enhanced our fraud detection and prevention capabilities, ” she explained. The Treasury Department attributed the success of AI to preventing and recovering over $4 billion in fraud overall during fiscal 2024, marking a six-fold increase from the previous year. US officials quietly began implementing AI for financial crime detection in late 2022, drawing from techniques already employed by many banks and credit card companies to thwart illegal activities. The initiative aims to safeguard taxpayer funds from fraud, which surged during the Covid-19 pandemic as the federal government rushed to distribute emergency aid to individuals and businesses. It’s important to note that Treasury is not utilizing generative AI, like those used in OpenAI’s ChatGPT and Google’s Gemini, which create images, compose music, and answer complex inquiries (despite sometimes struggling with simpler tasks). Instead, the fraud detection efforts rely on machine learning, a branch of AI adept at processing large datasets to make informed decisions and predictions based on acquired knowledge. AI proves to be invaluable in the battle against financial crime by efficiently analyzing vast arrays of data and identifying subtle patterns—much quicker than a human could. Experts assert that once advanced AI models are properly trained, they can uncover suspicious transactions in mere milliseconds. “Fraudsters excel at concealing their activities. They seek to exploit the system clandestinely, ” Miskell stated.

“AI and data utilization aid us in uncovering those concealed patterns and anomalies, allowing us to counteract them. ” This capability is particularly vital for the Treasury, which is one of the largest payers globally—if not the largest. Each year, the Treasury processes approximately 1. 4 billion payments totaling nearly $7 trillion to about 100 million individuals. It handles everything from Social Security and Medicaid payments to federal employee salaries, tax refunds, and stimulus checks. Given its critical role, the Treasury is an attractive target for fraudsters aiming to defraud taxpayers. Last year, the Internal Revenue Service indicated that it had implemented AI to identify tax evasion by analyzing extensive and complex returns from hedge funds, law firms, and other entities. Online payment fraud is projected to exceed $362 billion by 2028, according to Juniper Research estimates. Notably, some of this fraud is being exacerbated by AI technology itself. In a notorious incident earlier this year, Hong Kong police reported a finance employee was deceived by a deepfake video, leading to a $25 million payment to fraudsters. US officials have raised alarms about the new risks AI introduces into the financial system. In June, Treasury Secretary Janet Yellen warned bankers of the “significant risks” associated with AI in finance. Top regulators, led by Yellen, categorized AI late last year as an “emerging vulnerability” to the financial sector. Miskell emphasized that while AI systems will raise flags on suspicious transactions, a “human is always in the loop, ” as federal agencies ultimately decide whether something qualifies as fraud. The use of AI by the Treasury to combat financial crime is still in its early stages. Miskell mentioned that the Treasury is assessing how to incorporate the fraud-detection techniques utilized by leading banks and credit card companies, but refrained from providing specifics to avoid alerting potential fraudsters.