In 2025, CMS (Centers for Medicare and Medicaid Services) Deputy Administrator Kim Brandt announced that the agency had saved $2 billion since March 2025 using artificial intelligence to detect and prevent fraud in its health programs. CMS is the largest healthcare payer in the world, administering Medicare and Medicaid with a budget exceeding one trillion dollars annually. When an organization of that scale adopts AI-powered fraud detection and reports quantified results, the signal to the rest of the industry is clear: this approach works.
What CMS is doing
The CMS initiative, known as CRUSH (Center for Robotic Process Automation and Unified Solutions in Healthcare), uses artificial intelligence models to analyze claims before payment. The fundamental change is the timing of intervention: instead of paying first and auditing later, CMS evaluates claims against known fraud patterns and anomaly signals before authorizing disbursement.
This represents a structural transformation from the traditional "pay-and-chase" model that has characterized health program administration for decades. The previous model relied on post-payment investigations that, when successful, recovered a fraction of losses after months or years of litigation.
The $2 billion in savings reported by CMS is not a projection. It is not an estimate of potentially avoided fraud. It is documented savings, publicly announced by a senior agency official. This data point is significant because it establishes a verifiable precedent that pre-payment AI-based detection generates measurable returns at scale.
The state of Latin American insurers
While CMS implements AI fraud detection at scale, most health insurers in Latin America continue to rely on manual or semi-manual processes to audit claims. Audits are performed on samples, not on the entirety of transactions. Sample selection criteria are frequently static. And audit results are used to recover past losses, not to prevent future ones.
The magnitude of the problem in the region is documented. The Instituto de Estudos de Saude Suplementar (IESS) in collaboration with EY published a study in 2022 that estimated fraud and waste losses in Brazilian health plans at R$30-34 billion annually. This figure includes outright fraud but also operational waste — unnecessary treatments, duplications, and lack of adherence to clinical protocols.
Our own analyses in the region confirm this trend. 43.4% of pharmaceutical spend shows detectable anomalies, representing $5.1 million annually per 50,000 subscribers. The majority of these anomalies are not criminal fraud but systemic inefficiencies that a sample-based audit process will never capture in full.
The regulatory movement
It is not just CMS. Regulators in Latin America are moving toward data-based supervision. In Brazil, the Agencia Nacional de Saude Suplementar (ANS) has issued Resolucoes Normativas 656 through 659, which strengthen data reporting requirements and establish frameworks for more granular supervision of health plan operators. The regulatory direction is clear: regulators will demand more data, more transparency, and greater detection capability.
Insurers already implementing automated anomaly detection will be better positioned to meet these requirements. Those still relying on manual processes will face dual pressure: on one side, the operational losses they are not detecting; on the other, the regulatory requirements they will not be able to meet with their current processes.
What can be learned from CMS
There are concrete lessons from the CMS approach that apply directly to the Latin American context.
Pre-payment detection is structurally superior to post-payment audit. This is not an opinion. CMS tested both models for decades and is actively migrating to pre-payment. The reason is straightforward: it is cheaper to prevent a fraudulent payment than to try to recover it afterward.
Full coverage beats sampling. CMS analyzes the entirety of claims, not samples. When 100% of transactions are analyzed, coordinated fraud patterns hiding in volume become visible. Inspector AI operates on the same principle: every dispense event is evaluated against dozens of detection rules before authorization.
Savings are measurable from year one. The $2 billion CMS reported was within the first year of implementation at scale. This does not mean any insurer will see comparable savings — CMS operates at a scale that has no equivalent — but it does confirm that returns are measurable and do not require years of maturation to materialize.
What should not be extrapolated
It is important to be precise about what the CMS data demonstrates and what it does not. It demonstrates that AI fraud detection works at scale and generates quantifiable savings. It does not demonstrate that any implementation will generate a specific return, nor that CMS models are directly transferable to the regulatory and operational context of each Latin American country.
Insurers in the region operate under different regulatory frameworks, with different data structures and with fraud types that have local characteristics. The technology is transferable; the specific models must be calibrated to local context.
The time to act
The question for Latin American insurers is no longer whether AI fraud detection works. CMS has answered that question with $2 billion in evidence. The question now is how long it will take to adopt an approach that the world's largest healthcare payer has already validated.
Regulators are moving. Losses are documented. The technology exists and is proven. What is missing, in many cases, is the decision to implement.
To evaluate fraud and waste exposure in your claims data, request a free analysis.