Pandemic watchdog builds AI fraud prevention ‘engine’ trained on millions of COVID program claims
Pandemic watchdog builds AI fraud prevention ‘engine’ trained on millions of COVID program claims
Publish Date: 2026-01-13 19:06:00
Source Domain: federalnewsnetwork.com
- The Pandemic Response Accountability Committee (PRAC) has developed an AI-enabled “fraud prevention engine” trained on over 5 million applications for pandemic-era relief programs to identify fraudulent claims.
- PRAC’s fraud detection system can review 20,000 applications for federal funds per second and has allegedly recovered over $500 million in taxpayer funds.
- The PRAC is working to prevent fraud in programs funded by legislative initiatives, emphasizing the need for sustained data analytics capabilities to target organized, transnational criminals.
- The Government Accountability Office (GAO) estimated annual federal losses to fraud at $233 to $521 billion, with AI tools showing promise, though cautioning the need for high-quality data to avoid misleading results.
- The Treasury Department is directing agencies to access its Do Not Pay fraud prevention database to flag potential fraud before issuing payments, anticipating full utilization of all federal programs by the end of this fiscal year.
- To enhance Do Not Pay’s effectiveness, the Treasury aims to cross-check with authoritative federal databases like IRS and Social Security numbers before issuing payments, moving beyond a “trust but verify” approach.
