By Musheer Ahmed, PhD, Founder & CEO at Codoxo. Codoxo were finalists in the ‘Best Use of AI for Healthcare’ and ‘Most Innovative AI Technology’ awards at the 2025 AI Awards.

A New Kind of Fraud Problem

Generative artificial intelligence is accelerating transformation across healthcare, improving workflows, clinical documentation, and operational efficiency. At the same time, it is lowering the barrier to creating highly convincing fraudulent activity. What once required coordination, expertise, and time can now be generated quickly, inexpensively, and at scale. This creates a new problem for payers and investigators: not just more fraud, but fraud that is harder to recognize because it looks legitimate.

That matters in a system where the financial exposure is already enormous. U.S. healthcare spending reached approximately $5.3 trillion in 2024. Industry estimates suggest that up to 10% of that spend may be lost to fraud, waste, and abuse, representing hundreds of billions of dollars each year. At the same time, enforcement activity reinforces the scale and persistence of the issue. In a recent nationwide healthcare fraud enforcement action, federal authorities reported charges involving approximately $2.75B in intended losses, underscoring that fraud remains both widespread and highly adaptive.

For healthcare organizations, the challenge is no longer just identifying suspicious claims. It is understanding how fraud itself is changing and whether existing detection models are equipped to keep up.

Healthcare Fraud Is Becoming More Convincing

The most important change is not simply that fraud is increasing. It is that fraud can now be engineered to look legitimate.

Generative AI is accelerating this transformation along three dimensions: quality, accessibility, and scale. Modern AI systems can produce clinically plausible medical records that mirrors how providers actually write notes. They can do so without requiring deep medical expertise, while generating content rapidly and repeatedly, making it possible to scale healthcare fraud in ways that were previously impractical.

This introduces a new category of risk: synthetic credibility. Instead of relying solely on suspicious billing patterns, fraudulent activity can now be supported by documentation that appears internally consistent, well-structured, and aligned to the claim. Investigators may encounter fully AI-generated medical records, partially modified documentation layered onto legitimate records, or replicated narratives adapted across multiple patients.

In practice, these signals are already emerging. In one case, investigators reviewing a set of claims encountered supporting evidence that appeared highly detailed and clinically consistent across multiple patients. On closer inspection, the narratives followed nearly identical structures, with only minor variations that suggest a single source had been used to generate multiple records at scale. Each document appeared unique, but the underlying content had been replicated and adapted.

Similar patterns are now surfacing more broadly as organizations apply advanced analytics to uncover previously undetectable forms of fraud.

What makes these scenarios especially difficult is that the documentation can look polished and believable even when the underlying facts are not. In other words, the problem is no longer only the claim. It is the evidence attached to it.

Deepfakes have drawn attention as one visible example, particularly in imaging and identity verification. But the broader issue extends far beyond any single technique. It is the growing ability to manufacture clinical materials that fits seamlessly into existing workflows and passes traditional medical fraud review.

When Documentation Can’t Be Assumed True

This evolution exposes a fundamental limitation in how healthcare fraud has historically been investigated.

The traditional model is straightforward: identify suspicious billing patterns and known anomalies, request documentation, and validate the claim based on it. But that model assumes the record itself is reliable. Investigators are increasingly encountering cases where records appear unusually complete, well-organized, clinically coherent, and fully aligned to the claim under review. In some cases, these are the most polished records a team receives. Yet other signals may still link them to fraudulent activity. The more convincing the document, the harder it can be to challenge. Human reviewers alone are not well equipped to solve this. The result is a widening gap between how activity is executed and how it is detected.

Why Detection Must Evolve

This is where the industry’s approach to AI must mature.

For several years, AI adoption in healthcare fraud has been framed primarily as a tool for efficiency, helping teams process more cases, automate workflows, and accelerate investigations. Those gains are important, but they do not address the deeper shift now underway.

The next phase is about capability, not just speed. It requires applying AI across the fraud lifecycle, analyzing unstructured medical records at scale, detecting indicators of replication and manipulation, and connecting signals across claims, providers, and supporting records.

Generative AI plays a dual role in this model. While it enables new forms of fraud, it also enables more effective investigation. It can summarize medical records, align documentation to specific allegations, and surface key findings within the case itself. Tasks that once required days or weeks of manual effort can now be completed in hours, enabling SIU teams to review more cases with greater consistency.

Across the healthcare industry, these signals are emerging with increasing frequency, signaling a change that traditional approaches alone can no longer address. In real-world scenarios, teams applying AI-driven approaches have been able to accelerate investigations and uncover patterns that would have been difficult to detect through manual review alone.

From Reviewing Claims to Evaluating Evidence

The broader implication is that healthcare fraud strategy must expand beyond identifying suspicious claims to address the integrity of the evidence used to validate them.

This introduces a new dimension to fraud detection, one that focuses not just on what is submitted, but on how it was produced. In practice, this means moving beyond reviewing claims at face value to assessing whether the supporting evidence itself is authentic, consistent, and trustworthy.

This move represents a meaningful evolution in how fraud is approached. It also reflects a broader reality: as AI becomes embedded in healthcare workflows, not all AI-generated content will be problematic. The challenge will be distinguishing between legitimate, verifiable use of AI and content that is intentionally misleading or fraudulent.

That distinction will matter more as clinical records are produced at greater speed and scale, and as fraudsters learn to exploit the credibility of machine-generated content.

 The Next Move Is Ours

Healthcare fraud has entered a frightening new phase — one defined not just by scale, but by credibility.

Indeed, generative AI  advancements in healthcare save lives, time and cost. But for healthcare organizations, a new dark side is manifesting with fraudulent records that are convincing, scalable, and inexpensive to produce –– meaning, they are poised to exponentially rise. At the same time, the pressure on healthcare organizations to reduce waste, improve accuracy, and maintain trust will continue to grow.

The organizations that adapt successfully will not be those that simply automate existing workflows. They will be the ones  that rethink how fraud is detected and investigated, using AI to surface risk earlier, evaluating evidence more rigorously, and keeping pace with a rapidly evolving threat landscape.

Today, we’re operating within a system where fraud can be engineered to look real. As innovators in artificial intelligence, the imperative is equipping every healthcare organization to be capable of critically evaluating the evidence behind a claim, not just the claim itself.