Hearing Wrap Up: Congress Must Enact Solutions to Detect and Prevent Fraud in Federal Programs, Protect Taxpayer Dollars

WASHINGTON-Yesterday, the Subcommittee on Government Operations held a hearing on "Curbing Federal Fraud: Examining Innovative Tools to Detect and Prevent Fraud in Federal Programs." At the hearing, members examined innovative and cost-effective tools available to federal programs to detect and prevent improper and fraudulent payments, like the massive fraud recently uncovered in Minnesota's social services programs. Members also discussed how to effectively transition the Pandemic Response Accountability Committee (PRAC) from pandemic-focused efforts to government-wide fraud detection and prevention efforts.

Key Takeaways:

Current methods to detect and prevent fraud are ineffective and do not incentivize federal agencies to block improper payments.

• The current "pay and chase" method is ineffective. New, free, alternative tools will save American taxpayers billions and will ensure federal programs provide funding to legitimate recipients and beneficiaries.
• Ken Dieffenbach, Executive Director of the Pandemic Response Accountability Committee, testified that "As executive director of the PRAC, an entity Congress created to oversee over $5 trillion in relief funding, [I work] with a phenomenal team that is leveraging artificial intelligence to collect, organize, analyze data to rapidly provide insights into fraud risks. This proactive approach is clearly needed so that fraud is prevented before funds are dispersed. This past summer, the PRAC issued two fraud alerts that identified over $79 billion in potential fraud [in pandemic programs] that could have been prevented with pre-award vetting and cross-agency collaboration."
• Renata Miskell, Deputy Assistant Secretary for Accounting Policy & Financial Transparency at the U.S. Department of Treasury's Bureau of Fiscal Service, testified that "'Do Not Pay' is a government wide tool provided by [U.S. Department of the Treasury] for agencies and states operating federal programs to detect and prevent the leading causes of improper payments…Do Not Pay is a tool that helps these frontline workers detect risk when making awards and certifying payments. Despite its promise, Do Not Pay has faced two challenges. First, the program has been underutilized. In fiscal year [2024], only 4 percent of federal programs could access all available data. Second, Do Not Pay has not had sufficient authority to access key federal databases that could detect the most common drivers of improper payments, namely verifying identity, financial status and death."

The Government Accountability Office (GAO) estimates that the federal government loses hundreds of billions annually to fraud in federal programs.

• Sterling Thomas, Chief Scientist at the U.S. Government Accountability Office, testified that "[GAO] has a large body of work on fraud and improper payments in the federal government. We found the federal government reported an estimated $162 billion in payment errors or improper payments during fiscal year 2024, and that's almost certainly an underestimate, because it doesn't include estimates for some federal programs. We've outlined numerous ways that Congress and federal agencies can tackle this problem with existing capabilities, such as reducing data silos. For example, Congress could make permanent the social security administration's authority to share its death list with Treasury's Do Not Pay system. According to Treasury, just one year of access to this data resulted in total net benefits of $109 million, but Treasury's access to this data was set to expire in December."

Congress needs to develop a permanent, practical solution for the Pandemic Response Accountability Committee (PRAC) to maintain and expand their fraud detection, prevention efforts, and analytical capabilities.

• Mr. Dieffenbach testified that "To address these issues, the PRAC is developing an artificial intelligence enabled fraud prevention engine trained on 5 million pandemic applications and other data. The tool can quickly identify anomalies, trends, patterns and hidden connections in future applications before payments are made. Had our fraud prevention engine been in existence in March of 2020, pre-award vetting would have flagged at least tens of billions of dollars in fraudulent claims for further scrutiny, allowing agencies to prevent fraud. The PRAC is also actively engaging with partner inspectors general to identify opportunities to prevent fraud in programs funded in the One Big Beautiful Bill Act and to address fraud risks beyond identity theft and eligibility fraud issues or issues such as traditional procurement fraud. Of particular focus will be cross program risks, as fraudsters rarely target just one government program."

Member Highlights:

Subcommittee Chairman Pete Sessions (R-Texas) asked which federal agency controls the Federal Audit Clearinghouse (FAC) and how machine learning and AI can be used to catch funding inconsistencies in federal agencies.

Subcommittee Chairman Sessions: "I'd like for you, Dr. Thomas, to not recreate what we talked about yesterday in my office, but pretty close to that about how important these exercises are on a really program-by-program basis…to look at inconsistencies that would draw you to those things. Do you mind taking a minute, Mr. Mfume? I think would learn a lot from this. Perhaps he knows it, but hearing from you, Dr. Thomas, would be important."

Dr. Thomas: "Yeah, I appreciate that. So we have a demonstration program that I talked about earlier that we call 'Facet' that is really built for, just as Chairman Sessions talked about, identifying indicators of potential bad behavior, really just data points in a large data collection. And this is the Federal Audit Clearinghouse that, you know, can be used across programs. So I'll give you a couple of thoughts there. The Federal Audit Clearinghouse, you know, houses over $1 trillion of spending every year. And these are audits that are done not only on federal programs, but these are state programs that we're using. Federal money and local programs are using federal money. We use a combination of AI, natural language processing, and machine learning to identify, you know, one pattern here say that, you know, is in a food support program that then is replicated in another program, completely different and a completely, completely different locality. These indicators of fraud don't mean fraud is happening, but it means someone should take a look at it. And that's really the message we try to put people is that not only should they look at it, they should understand is it actually a problem with the program. Is it a problem with the way the audit was done? Or was it a problem where the design of the program isn't actually collecting the right piece of data, and then you can help that program say, okay, 'all new grantees should produce this piece of data, which will then go into your fraud risk model."

[…]

Subcommittee Chairman Sessions: "Dr. Thomas, which agency manages the Federal Audit Clearinghouse?"

Dr. Thomas: "So the federal audit clearinghouse is actually the responsibility of the Office of Management and Budget. They delegate that to the General Services Administration, and that is actually who's operating it. And when we access it, we work with them to work through their APIs. Now, they've had some challenges with staffing recently, which has delayed our ability to make some updates. I'll give you a specific example. Several of their newer APIs were mislabeled on what versions they were, so it made the system not work appropriately. But that is, you know, something that could be improved. And so that's where it lives."

Subcommittee Chairman Sessions: "So what would you say is the current status recognizing, as you have alluded to, the delay and perhaps the misrepresentation of the data? Where does that stand today and who has the responsibility on your side at GAO and within the agency?"

Dr. Thomas: "So the agency doing the work is the GSA. The responsibilities that it, [Office of Management and Budget], within our group-is our innovation lab that is working with them to identify these errors and improve them. Now, we have recently been working on getting abetter response from them. But I think it gets back to my earlier statement of, you know, just enabling them with a, you know, a workforce that has the technology skills to run an important program like this. And there's over $1 trillion a year in spending in the federal audit clearinghouse. It's a rich data set, if we work to improve it and extract value out of it."

Subcommittee Chairman Sessions: "So that's what's at risk, that trillion dollars? I'm not trying to give the whole example of a trillion [lost] , but that is how big of a problem this would help solve."

Dr. Thomas: "That's how much is obligated within the federal audit clearinghouse database. Yeah, those that's what the audits represent."

Subcommittee Chairman Sessions: "Well, I respect that. And we will take that up. Also, I think that this subcommittee is very capable of looking at things. And if $1 trillion not big enough, we're in the wrong business."

Rep. Gary Palmer (R-Ala.) inquired about the PRAC's progress to detect and prevent fraud going forward.

Rep. Palmer: "I'm sad to say that when I came into Congress, our improper payments were running somewhere around $130 to $150 billion a year. It has since surpassed that considerably. During COVID, it got even worse, particularly with the payroll subsidies that we sent out. The additional unemployment insurance, the fraud, was massive. On that end, I think, one state in particular, in the first four months, sent out about $1 billion in fraudulent payments. I know that we're working on that, but I want to know what progress has been made in that area to address the fraud, uh, from the pandemic funding."

Mr. Dieffenbach: "Thank you, Congressman Palmer. We've done a tremendous body of work on identifying what went wrong during the pandemic. We've issued lessons learned reports and a blueprint for program integrity about how to ensure this never happens again. But let me answer your question quickly with a story that Chair Sessions I know will appreciate. One of the many projects we did was we examined recipients of [Department of Housing and Urban Development] low-income housing benefits with the Social Security Numbers that were also used to obtain Small Business Administration (SBA) PBP loans. So, folks that were claiming a low income to get the housing benefit and a high income to get the PPP loans often-forgivable loans-we found 40,000 instances in which the disparity in income between those two programs was ten times or greater, and that impacted $860 million in PPP loans. So I think telling those stories is important. That's what one of many, many risks and programs. And the end to that story is that ID theft or lies occurred in the HUD program or the SBA program or both, and legitimate victims that needed housing or PPP loans didn't get them."

Rep. Brian Jack (R-Ga.) asked what would happen if the PRAC had expired, as well as how AI can be used to root out waste, fraud, and abuse in federal programs.

Rep. Jack: "Could you walk us through what the world would look like if PRAC had expired, if we no longer had access to this great, you know, great tool that is meant to root out waste, fraud and abuse. Walk us through what life would look like if this had expired and we had not been able to extend it last year."

Mr. Dieffenbach: "Great question, Congressman Jack, and for your support. The PRAC has assembled a phenomenally unique set of data about pandemic fraud, about program fraud in general, about the patterns, the trends, the anomalies. We've issued a number of alerts. So had we expired, the ability to provide the insights I just spoke to a minute ago would be gone. There would be another disaster. [Congress] would fund an emergency data analytics capability, and we would have to spend a year or two to rebuild that. So we have been able to continue to keep pace. The fraudsters don't take naps or take breaks. So we've been able to continue to build upon everything we've learned over the last six years. And I think it's been a tremendous asset to Congress and to the taxpayer."

[…]

Rep. Jack: "I'm fascinated by the utility [AI] could offer in this space. I welcome any closing thoughts from you on how best AI could continue to strengthen our mission to eliminate waste, fraud and abuse?"

Dr. Thomas: "Yeah, there's like I said, there's tremendous opportunity in AI. I think the foundational components, that actually Ms. Miskell was talking about, are critically important, and that is build a solid data collection of what is AI or what is fraud, so that AI can learn. The challenge with AI is that it doesn't know the difference if you don't tell it. And so if we properly label all of these examples that both of my colleagues are finding, collect the data that represents this is fraud, we can then start to train a tool to do this broadly. And then all of a sudden you can leverage without having to have the analysts read all of these audits and go through the data by themselves. They can be empowered with this tool that can pull [information] out. 'These are the indicators of fraud. These are the patterns that just don't represent typical behavior of a Pay ID in this program.' And now someone can go look at it. That is only possible if we have this foundational kind of gold standard of this is fraud database."

Click here to watch the hearing.