A fraud-detection AI model trained on COVID-19 loan data could have flagged potentially tens of billions of dollars in payments before they went out, reducing the feds' pay-and-chase cleanup, the US government's Pandemic Response Accountability Committee told Congress on Tuesday.

Hindsight is 20/20, of course, and the roughly five million Small Business Administration COVID-19 Economic Injury Disaster Loan applications used to train PRAC's "Fraud Prevention Engine" were only available after the programs were already up and running, by which point billions of dollars had been paid out and later identified as potentially fraudulent.

Regardless, PRAC executive director Ken Dieffenbach told the House Oversight and Government Reform Committee at a Tuesday [1]hearing that it's time to put those lessons to work to prevent other cases of government fraud, something that the Trump administration and Republican congressional leaders [2]ostensibly care a great deal about.

[3]

"The time is now to use this data to prevent fraud schemes before taxpayer dollars are lost and hold wrongdoers accountable," Dieffenbach told the Committee.

[4]

[5]

According to the PRAC director, the Fraud Prevention Engine was developed as a proof of concept "to determine if such a model could be developed and to identify and address any technical hurdles." By Dieffenbach's account, it appears the project was a success.

The Engine was built with a bunch of modular components, including unsupervised machine learning models used to detect anomalies, supervised ML that identifies patterns consistent with pandemic fraud cases, and rules-based flags that catch invalid Social Security and employer identification numbers. According to Dieffenbach, such small anomalies can often identify hidden connections, like shared bank account numbers among supposedly independent applicants, that suggest fraud.

[6]

And it can do it quickly: Dieffenbach said the AI can process 20,000 applications per second.

"We estimate that had our Fraud Prevention Engine been in use in March 2020 it would have quickly … flagged potentially tens of billions of dollars in payments for further scrutiny before the funds were disbursed," Dieffenbach told the Oversight Committee.

PRAC says it has helped recover just over $500 million so far, a small fraction of the payments Dieffenbach says the Engine could have flagged before disbursement.

[7]Biotech exec sentenced to eight years for COVID-19 testing finger-stick fraud

[8]Ex-Microsoft exec admits he tried to swindle $5.5m out of taxpayers with COVID-19 relief loans for bogus biz

[9]Billions lost to fraud and error during UK's pandemic spending spree

[10]GAO finds billions in possible government savings, all without Elon's help

In other words, there's a lot of work to be done, and that's just on pandemic fraud. Dieffenbach isn't stopping the Fraud Prevention Engine's work at PRAC's COVID loan recovery efforts, though.

Trump's budget reconciliation bill [11]passed last summer extended PRAC's mandate to cover fraud detection and prevention in programs funded under the law, pushed the Committee's end date out to September 30, 2034, and provided $88 million in funding to support its expanded oversight work.

[12]

Since then, PRAC has begun working with Inspectors General across multiple agencies to deploy the Fraud Prevention Engine and related analytics in support of oversight of programs funded by the reconciliation law. Dieffenbach said the effort has already helped identify specific programs where PRAC's data and technology can provide valuable insights.

"We have identified several specific programs where our data and technology can provide valuable insights," the director said.

Pete Sessions (R-TX), a member of the Oversight Committee, says the Engine needs to find a permanent home before PRAC disbands in eight years.

"A permanent solution that maintains the analytic capacities and capabilities that have been built over the past six years is necessary and needed," Sessions said. "Its database is billions of records deep and it has begun to pay for itself."

We reached out to PRAC to learn if it's been in discussions to find a permanent home for the Engine, but didn't hear back. ®

Get our [13]Tech Resources



[1] https://oversight.house.gov/hearing/curbing-federal-fraud-examining-innovative-tools-to-detect-and-prevent-fraud-in-federal-programs/

[2] https://www.theregister.com/2025/04/20/musks_doge_promises_fail/

[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/publicsector&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aWlxnVep7AKPD7pP5gdKCAAAAAU&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/publicsector&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aWlxnVep7AKPD7pP5gdKCAAAAAU&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/publicsector&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aWlxnVep7AKPD7pP5gdKCAAAAAU&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/publicsector&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aWlxnVep7AKPD7pP5gdKCAAAAAU&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[7] https://www.theregister.com/2023/10/19/biotech_exec_sentenced_for_covid19/

[8] https://www.theregister.com/2021/03/16/ppp_fraud/

[9] https://www.theregister.com/2024/02/09/nao_pandemic_fraud_data/

[10] https://www.theregister.com/2025/05/14/billions_in_savings_possible_in/

[11] https://www.theregister.com/2025/07/02/space_shuttle_relocation_bill/

[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/publicsector&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aWlxnVep7AKPD7pP5gdKCAAAAAU&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[13] https://whitepapers.theregister.com/