The U.S. Department of the Treasury says its expanded use of machine learning systems helped detect and prevent billions of dollars in fraudulent payments in 2024.
The treasury is the check-writer for many federal programs and annually processes around 1.4 billion payments worth $6.9 trillion for programs like Social Security and Medicaid.
During the most recent fiscal year, which ended in September, the agency’s new data-driven approach to rooting out bad actors contributed to the prevention and recovery of more than $4 billion in fraudulent payments, according to a press release. That’s a more than sixfold increase over the $652.7 million in fraudulent payments detected or recovered during the 2023 fiscal year.
The agency credited the increase to its new data-driven approach to fraud detection. That includes using machine learning to identify instances of fraud and to prioritize high-risk transactions for further investigation. The Treasury has also partnered with other federal and state agencies to share information through its Do Not Pay database and other payment integrity tools.
“Treasury takes seriously our responsibility to serve as effective stewards of taxpayer money. Helping ensure that agencies pay the right person, in the right amount, at the right time is central to our efforts,” Treasury Deputy Secretary Wally Adeyemo said in a statement. “We’ve made significant progress during the past year in preventing over $4 billion in fraudulent and improper payments. We will continue to partner with others in the federal government to equip them with the necessary tools, data, and expertise they need to stop improper payments and fraud.”
While $4 billion in fraudulent payments prevented or recovered is no small amount, it pales in comparison to the government’s estimates of how much fraud occurs.
In April, the federal Government Accountability Office estimated that federal agencies lose between $233 billion and $521 billion annually to fraud. The GAO report recommended that the Treasury, due to its central role in processing payments, should better leverage data analytics tools.
Both government agencies and financial institutions have increasingly come to rely on machine learning algorithms to identify fraudulent actors. These systems use a wide range of data about payment recipients—including details about their bank accounts, physical addresses, IP addresses, demographic information, usernames, and passwords—to identify patterns linked with fraud.
As the Treasury has noted in previous reports about financial sector fraud, that kind of “historical data used to train fraud-detection models could contain biases, such as the overrepresentation of certain demographics in anti-fraud cases.”
Read the full article here
