By Dr. Musheer Ahmed, CEO, Codoxo
LinkedIn: Musheer Ahmed
LinkedIn: Codoxo
A regional health plan opens an investigation after finding unusual billing patterns from a high-volume orthopedic provider. The investigators review several previously paid lumbar spinal fusion claims, including physician notes, treatment histories, and diagnostic imaging. At first, the claims appear well supported, with complete documentation and imaging consistent with the diagnoses.
Upon closer review, new questions are raised. The metadata does not match the imaging device on file. Some timestamps are inconsistent, and parts of the images show signs that they may have been modified outside the original acquisition process. No single finding necessarily proves fraud. However, taken together, they give investigators a reason to question something payers have historically been able to take largely for granted: whether the medical evidence itself is authentic.
That is an emerging blind spot for healthcare payment integrity teams. Most fraud controls were built to determine whether a claim is consistent with the evidence supporting it, but Generative AI has created a new way to game the system. The claim, physician narrative, and diagnostic evidence can all agree with one another when the supporting evidence has also been manipulated or cloned to fit the bill.
Fraud detection has focused on claims, not provenance
Healthcare fraud is already a widespread and costly problem. The FBI estimates it causes tens of billions of dollars in losses each year. Federal enforcement actions illustrate the sheer scale of the problem. In June 2026, the Department of Justice’s National Health Care Fraud Takedown resulted in charges against 455 defendants connected with more than $6.5 billion in alleged fraud.
Payers have responded by building increasingly sophisticated fraud, waste, and abuse programs. These systems work by analyzing coding patterns, provider behavior, claim timing, and relationships across claims datasets, looking for behavior that doesn’t fit an expected pattern. Medical records then help validate what the data suggests: claims analytics flag unusual utilization or coding, clinical reviewers determine whether documentation supports medical necessity, and investigators compare the services billed with physician notes, tests, and imaging. Those processes work well when the supporting evidence is genuine.
What they don’t necessarily establish is whether a record actually came from the clinical encounter it claims to represent. A CT scan may look medically plausible, a physician’s note may contain the expected terminology, and a treatment history may align closely with the diagnosis and procedure, all without any of it reflecting what actually happened to the patient.
Generative AI makes it easier to manufacture that consistency where it might not clinically exist.
The problem extends well beyond healthcare. Organizations in many sectors now face manipulated images, synthetic identities, cloned voices, and other forms of AI-generated evidence. Research supported by the Royal Society illustrates the difficulty of relying on human review alone. In one experiment, only 10.7% of participants correctly identified a deepfake without a warning. Even after researchers explicitly told participants that one of the videos was a deepfake, only 21.6% identified it correctly.
For payers, that broader authenticity problem now intersects directly with payment decisions.
Medical imaging raises the stakes
This issue applies to many forms of medical documentation, but diagnostic imaging deserves particular attention. CT and MRI studies often carry significant weight in decisions about medical necessity, since they can support expensive procedures, specialty referrals, and treatment plans. They also contain technical information that investigators may not review as part of a standard claims workflow.
As image generation and manipulation tools improve, payers may encounter altered studies that still look highly credible on routine review. An image could be manipulated slightly to support pathology that did not exist, exaggerate the severity of a condition, or make a high-cost procedure appear medically necessary.
The point isn’t that payers should treat every unusual image as suspicious, since imaging files can contain irregular metadata or technical artifacts for legitimate reasons. Rather, a pattern of authenticity signals should flag a claim as requiring additional investigation.
In the spinal fusion example, an investigator might confirm the originating imaging system, compare timestamps with other records, review related studies, or request documentation directly from the imaging provider. These checks give investigators another source of evidence before they make a determination.
The financial case for earlier detection is getting stronger
This risk is emerging while payer economics are already under significant pressure. U.S. healthcare spending reached $5.3 trillion in 2024, up 7.2% from the prior year, according to CMS. Private health insurance spending rose 8.8% to $1.64 trillion.
Fraud prevention alone will not solve medical cost inflation, but avoidable improper payments add to that pressure. High-cost procedures illustrate the potential exposure.
A 2026 JAMA Network Open analysis found that the mean inpatient cost for lumbar fusion reached $45,458 in 2023. Costs ranged from $33,610 for a one-level single-column fusion to $55,034 for a multilevel anterior-posterior fusion. Commercial reimbursement can vary further based on geography, payer contracts, implants, and surgical approach.
The exposure can become meaningful even at relatively low volumes. Consider a hypothetical mid-sized health plan that processes 20 to 25 lumbar fusion procedures in a year. If just one to three of those cases involved manipulated supporting documentation, and each represented $30,000 to $80,000 in payments, the potential annual exposure would range from roughly $30,000 to $240,000. (Note: this is an illustrative calculation, not an observed rate of manipulated imaging.)
The cost doesn’t end with the original claim, either. Once a payer identifies a questionable payment, resolving it may require investigators, clinicians, provider outreach, audits, appeals, and recovery efforts. More sophisticated supporting documentation can make that work even harder, since investigators first have to figure out which parts of the record they can actually trust.
This is why synthetic medical documentation has a ripple effect across SIU, payment integrity, compliance, clinical operations, finance, and actuarial teams, each of which has a stake in catching improper payments before they become expensive recovery projects.
The pressure is also broader than commercial health plans. Payers, pharmacy benefit managers, and state and federal healthcare agencies face different types of exposure, but all are under growing scrutiny to demonstrate effective fraud prevention, payment integrity, and oversight. As AI-generated documentation becomes easier to produce, readiness for this category of risk will increasingly fall under that same mandate.
Authentication is becoming part of payment integrity
Health IT leaders should start by identifying where their organizations currently assume documentation is authentic. Which payment decisions rely heavily on uploaded images or medical records? Which procedures combine high reimbursement with documentation that could be manipulated? Where can investigators check provenance today, and where are they working primarily from the record they receive?
Organizations can then layer authenticity signals onto the workflows they already use, drawing on indicators like file provenance, metadata inconsistencies, evidence of image manipulation, unusual document creation patterns, or discrepancies between submitted records and trusted source systems.
The goal isn’t to create another review queue, but to help existing payment integrity and fraud teams determine where deeper investigation is warranted.
Healthcare organizations have spent years improving their ability to determine whether claims are supported by clinical evidence. AI now adds a second question: can they trust that evidence?
As synthetic medical content becomes cheaper and more convincing, that question will get harder to ignore. Payers that address it early will be able to strengthen fraud prevention while also limiting the need for additional manual review and pay-and-chase activity.
Authentication is becoming part of payment integrity, and health plans should design for that reality now.