Federal fraud in the US: CMS and Education put the savings from stopping payments before they go out in the billions

By Federal News Network · November 6, 2026.

The U.S. Centers for Medicare & Medicaid Services (CMS) says its Center for Program Integrity achieved $42 billion in annual savings, against spending of about $2 trillion a year on Medicare and Medicaid. That is roughly 2% of the total. The figure came from Bethany Messick, deputy director of the center's analytics and systems group, during a NextGov webinar on November 5.

The Department of Education offered another data point. Richard Lucas, acting chief operating officer of the Federal Student Aid office, said a real-time detection tool on the FAFSA form (the application for student aid) is on track to prevent about $1.2 billion in fraud. According to Education, the tool was rolled out this year with an unnamed “leading financial services firm.”

Here is how it works, according to the agencies themselves. CMS's fraud prevention system uses more than 300 models that continuously monitor providers and flag behavior that is “outside the norm.” Investigators then review those alerts. If there is a credible allegation of fraud, CMS can suspend payment while it investigates and, if the fraud is confirmed, remove the provider from the programs. On the FAFSA, the system assesses risk while the family fills out the form. Most applicants move through without friction, and a minority must complete extra steps to verify their identity.

Messick said CMS's fraud-focused centers already use AI tools to quickly block suspected fraud. She also stressed that the final decision on a payment still rests with a human analyst, because AI “can hallucinate.” Lucas added that Education is seeing attacks shift from a single vector to several, often targeting multiple programs at once.

Our take. The underlying shift is the move away from “pay-and-chase” (paying first and pursuing later) toward stopping the money before it goes out. That is a genuine improvement in logic: recovering a fraudulent payment that has already been collected is usually far harder than not making it in the first place. It also fits a thesis we hold: the near-term risk of agentic AI is the automation of fraud, and the sensible response is a defense that is automated too.

Even so, these figures should be read with caution. They are self-reported, and the article does not explain how “savings” are calculated or how much reflects confirmed fraud versus payments avoided on suspicion. The Education figure is also a projection (“on track to”), not a final result. Nor do we know how many legitimate applicants face friction or delays because of false positives, which is the hidden cost of any system that blocks first and investigates later.

That is why it matters that CMS keeps a person accountable for the final decision. When an algorithm can suspend a payment to a hospital or block a student's financial aid, accountability and avenues for appeal matter as much as the detection rate. More broadly, the governance of these systems, with auditing and a right to respond, will be the real differentiator between AI that protects public money and AI that produces injustice at scale.

In the long run, the lesson is optimistic but demanding. If AI can make public services more efficient and less exposed to fraud, there are more resources for healthcare and education. But that abundance only materializes if the systems are measurable, auditable and fair to those who have done nothing wrong.