Artificial intelligence is moving deeper into hospitals, clinics and insurance workflows, promising faster documentation, more efficient analysis and better support for clinicians. But new research from the Blue Cross Blue Shield Association is raising a costly question: what happens when AI changes the way medical conditions are recorded and billed without a corresponding change in the care patients actually receive? An analysis released in September 2026 estimates that changes associated with AI supported hospital coding added about $942 million in healthcare spending for Blue Cross Blue Shield companies between 2023 and 2025. The finding does not establish that diagnostic AI itself caused those costs, but it highlights a growing concern across healthcare: unverified or poorly governed automation can create financial consequences far beyond the software screen where the change begins.
Blue Cross Analysis Points to a Disconnect Between Coding and Care
The latest analysis from the Blue Cross Blue Shield Association examined claims data and found a significant increase in patients being documented as medically complex. The share of cases classified as medically complex rose from about 37 percent at the beginning of 2023 to 40 percent by the end of 2025. Researchers estimated that this shift was associated with roughly $942 million in additional spending for Blue Cross companies.
The central issue is not simply that artificial intelligence is being used. Hospitals have increasingly adopted automated tools that can review medical records, identify possible diagnoses and assist with the process of assigning billing codes. The concern arises when an automated system identifies additional conditions that place a patient into a higher reimbursement category even though the treatment delivered does not appear to have changed in the same way.
Blue Cross researchers found that many of the additional diagnoses involved secondary conditions. These conditions can be identified through individual laboratory results or other information contained in a medical record. Such information can be particularly easy for automated systems to detect at scale. Once entered into the record, however, a diagnosis can affect how an entire episode of care is classified and reimbursed.
For a patient sitting in a hospital room, this process may be invisible. The nurse still checks vital signs, the physician still reviews the chart and the patient still receives the same treatment. Yet behind the scenes, an additional diagnosis can change the financial category assigned to that hospital stay. When that pattern is repeated across thousands of cases, even a relatively small difference per claim can become a substantial national expense.
The Findings Are About AI Enabled Coding, Not Proof That Diagnostic AI Causes Higher Costs
The distinction matters. The Blue Cross analysis focuses on artificial intelligence assisted hospital coding and claims patterns rather than proving that AI diagnostic systems are independently responsible for the increase. That difference is especially important as hospitals experiment with medical imaging systems, clinical decision support software, generative AI documentation tools and automated diagnostic platforms.
The broader concern is that an AI system can influence a medical workflow without making the final clinical decision itself. A tool may identify a possible condition, suggest language for a medical record or flag information that a coding system later uses. Human professionals may then review the output, accept it or incorporate it into a claim. The resulting financial impact can therefore involve several steps rather than a single automated decision.
BCBSA previously reported a related analysis in March 2026 involving tens of thousands of maternity admissions. Researchers identified a sharp rise in claims containing acute posthemorrhagic anemia, a condition that can involve substantial blood loss and often requires treatment such as a transfusion. The analysis found that many patients receiving the diagnosis did not receive the expected treatment. Researchers estimated that aggressive AI enabled coding practices could be associated with about $2.3 billion in inpatient and outpatient spending nationwide.
Those figures are estimates, and the research does not establish that every additional diagnosis was incorrect. Medical records can contain legitimate conditions that do not require the most intensive treatment. Nevertheless, the pattern has prompted questions about whether automated coding systems are identifying clinically relevant information or simply finding more opportunities to classify patients at higher levels of severity.
Why Diagnostic AI Requires More Than a Promising Accuracy Number
Medical AI is often discussed through the language of accuracy. A diagnostic system may perform well in a controlled study, identify abnormalities quickly or match specialist performance on a specific task. Real healthcare is more complicated. Patients differ by age, geography, medical history, socioeconomic circumstances and access to care. A system trained on one population can perform differently when introduced into another.
The World Health Organization has repeatedly called for stronger oversight of artificial intelligence in health. Its 2026 work on AI related health research stresses the need for ethics review, accountability and safeguards because existing oversight systems may not address all of the risks created by AI. The organization also identifies concerns involving biased data, transparency, equity and responsibility when systems are used in real healthcare environments.
WHO has also urged health systems to keep human judgment at the center of AI supported decisions. Its guidance on evidence informed health policy recommends technology readiness reviews before deployment and ongoing human verification after implementation. The principle is straightforward: artificial intelligence can process information rapidly, but clinicians and health administrators still need to determine whether the information is appropriate for a particular patient and situation.
For hospitals, this means that buying an AI tool is only the beginning. A responsible implementation requires testing the system against local patient populations, monitoring errors, documenting how recommendations are reviewed and establishing clear responsibility when an automated output is wrong.
How Higher AI Related Costs Can Reach Patients and Families
Healthcare costs rarely remain confined to the organization that first incurs them. When insurers receive higher claims, the financial effects can eventually appear through premiums, employer sponsored coverage, government spending and out of pocket expenses.
That makes the Blue Cross findings relevant even to people who never knowingly interact with an AI medical system. A patient may visit a hospital for a routine procedure and have no idea that automated software helped analyze the chart. Yet if coding practices consistently classify similar episodes of care as more complex, the financial consequences can spread across an insurance population.
For families already balancing medical bills, prescription costs and insurance premiums, these changes can feel frustratingly distant from the technology that caused them. The software may operate behind a computer screen, but the resulting expense can arrive in a very tangible form through a premium notice or a medical bill.
Hospitals Need Clear Rules for AI Assisted Workflows
The growing use of AI does not mean healthcare systems should abandon automation. Properly tested systems can reduce repetitive administrative work, help clinicians organize large volumes of information and support research. The challenge is establishing boundaries that separate useful assistance from unchecked automation.
Health systems evaluating AI tools should pay particular attention to several areas:
- Whether the system has been independently evaluated using patient populations similar to those served by the hospital.
- Whether clinicians can review and challenge AI generated diagnoses, documentation or coding suggestions.
- Whether changes in billing patterns are monitored after an AI system is introduced.
- Whether the organization can trace how an AI recommendation affected a medical record or insurance claim.
- Whether patients and clinicians have clear information about when automated systems are influencing healthcare decisions.
These safeguards are not merely technical requirements. They are part of maintaining trust between patients, doctors, insurers and healthcare organizations. When a system changes a diagnosis or financial classification, someone must remain accountable for checking whether that change makes clinical sense.
The Global Healthcare Debate Is Moving From Innovation to Evidence
The Blue Cross findings arrive as health systems around the world are moving from experiments with artificial intelligence toward broader deployment. WHO has reported that AI is already being used for diagnostics in many countries, while governance frameworks and liability rules remain less developed in numerous health systems.
That gap is particularly significant in lower resource settings. A tool that promises to compensate for shortages of specialists can be attractive where access to advanced medical expertise is limited. Yet those same systems may have fewer resources for validation, monitoring and technical oversight. Poor quality data or an algorithm that performs differently across populations can therefore create risks that are difficult to detect.
WHO’s recent work on responsible AI in health points toward a model in which evidence, safety, transparency and equity are evaluated alongside technical performance. More information about the organization’s current approach to artificial intelligence in healthcare is available through its global health AI resources.
What Comes Next for Insurers and Healthcare Providers
Insurance companies are likely to continue examining claims data for unusual changes as AI becomes more deeply integrated into hospital administration. Insurers have access to large populations of claims, giving them an opportunity to identify patterns that may not be visible inside an individual hospital.
Hospitals, meanwhile, will need to demonstrate that AI supported documentation and coding reflect the care actually provided. That does not require rejecting automation. It requires measuring its effects after deployment rather than assuming that a system is beneficial simply because it reduces the amount of manual work.
The Blue Cross research also gives healthcare leaders a practical financial lesson. An AI tool can appear inexpensive when evaluated as a software subscription or implementation project, while its broader economic effect may emerge through thousands of altered claims. The real cost of an automated system therefore includes not only licensing and infrastructure, but also oversight, validation, auditing, training and correction when the technology produces unreliable results.
A More Careful Path for Medical AI
We are entering a period in which artificial intelligence may become a routine part of how healthcare information is collected, interpreted and paid for. That future can bring meaningful benefits, but only if hospitals and insurers measure what happens after the software is switched on.
The latest Blue Cross findings provide a useful warning without proving that AI itself is inherently inflationary. The evidence points to a narrower and more practical issue: when automated systems alter medical coding or documentation, healthcare organizations need to verify that those changes correspond to real clinical circumstances.
For patients, the most important principle is simple. Technology should support better care without making the healthcare system harder to understand or more expensive without a clear clinical reason. As AI moves from pilot programs into everyday medical workflows, the strongest measure of success will not be how quickly a system produces an answer. It will be whether that answer is accurate, clinically meaningful, fairly applied and connected to the care a patient actually receives.
Further background on the latest Blue Cross analysis and its findings is available through Blue Cross Blue Shield Association research and health policy resources.
