Insurers claim AI is already increasing healthcare costs

When I first read this Blue Cross Blue Shield report, my gut reaction was a strange mix of deep frustration and quiet amusement. We have been sold a utopian narrative for three straight years that artificial intelligence would automatically slash administrative overhead, eliminate operational waste, and make medical care dirt cheap for everyday people. The tech industry pitched AI as the ultimate deflationary engine for healthcare. Instead, we just got hit with a cold, $942 million reality check. Blue Cross Blue Shield claims that hospital use of AI diagnostic and billing tools racked up nearly a billion dollars in additional healthcare spending across just twenty-four months. I've been following this space for a while now, and let me be direct about this: blaming artificial intelligence for driving up costs misses the entire economic structure of modern medicine. The technology isn't broken. It is doing exactly what our broken financial incentives trained it to do.

Key Takeaways

• AI is currently optimized for revenue capture: Hospitals are using natural language processing and clinical documentation software to capture every conceivable billable code, driving systematic "upcoding."
• Efficiency does not mean cost reduction: In a fee-for-service environment, saving clinicians time simply creates room to order more high-margin diagnostic tests and billable procedures.
• An algorithmic arms race is underway: While hospitals deploy machine learning to maximize billing reimbursements, insurance providers use automated software to deny claims, creating massive friction.
• Regulatory reform must target incentives: Technology will only lower costs if healthcare shifts away from fee-for-service and toward outcome-based, capitated payment models.

The Mechanics of Algorithmic Upcoding

To understand where that $942 million went, you have to look closely at how hospital billing infrastructure actually operates. Healthcare providers do not simply send an invoice for "treating a sick patient." They rely on incredibly complex coding systems like ICD-10 and CPT to categorize every single action, symptom, and pre-existing condition. For decades, human medical coders combed through handwritten charts and electronic health records to translate clinical notes into billable events. Humans miss things. They get tired, they run out of time, and they overlook subtle secondary conditions that could legally bump a patient into a higher complexity tier.

Enter specialized clinical documentation improvement tools powered by large language models and natural language processing. These software platforms scan doctor notes in real time, constantly poking physicians to add specific terminology. If a clinician writes that a patient had low blood oxygen, the AI instantly prompts: "Did you mean acute hypoxic respiratory failure?" That single phrase transformation elevates a basic diagnostic-related group code into a major complication category. It instantly adds thousands of dollars to the final insurance claim.

Here's what gets me about this whole setup: none of this is technically illegal. The patient actually had low oxygen, and the medical threshold for that diagnosis was technically met. But before AI, that level of granular documentation was humanly impossible to sustain across millions of patient encounters. Now, algorithms act as tireless revenue miners, sifting through millions of medical charts to extract every micro-cent of potential billing. Hospitals call it "revenue cycle optimization." Insurers call it systematic inflation. I call it machine-learning-assisted upcoding.

The Corporate Narrative vs. Healthcare Reality

When venture capital firms poured billions into healthtech over the past five years, the narrative was remarkably clean. Silicon Valley promised that automating clinical charting and diagnostic triage would eliminate physician burnout and pass operational savings down to payers and consumers. But that promise assumed a free-market system where efficiency leads to lower prices. Healthcare economics do not work like consumer electronics. When microchips get faster to produce, television prices fall. When hospitals become more efficient, they simply increase their throughput of billable events.

If an AI medical scribe saves an emergency room physician two hours of administrative work per shift, that doctor does not go home early. The hospital fills those saved hours with four additional patient visits, three extra blood panels, and two advanced imaging scans. In a traditional fee-for-service framework, saved time is immediately monetized into additional clinical volume. Furthermore, diagnostic algorithms trained to detect subtle anomalies often trigger a wave of defensive medicine. When a machine highlights a 0.5% risk of a rare vascular issue on an X-ray, no sane physician is going to ignore the software warning and risk a malpractice lawsuit. They order an expensive follow-up MRI.

This dynamic raises massive questions about whether the software boom in enterprise healthcare is actually sustainable. We have to ask ourselves: Is the AI industry really ready to slow down and confront the mess it is creating, or will platforms keep pushing rapid deployment regardless of downstream financial consequences? In my view, tech vendors sold software as a cost-cutting tool while secretly pitching it to hospital CFOs as a high-margin revenue multiplier. Both things cannot be true at the same time.

The Insurer vs. Hospital Algorithmic Escalation

There is a deep hypocrisy at the center of Blue Cross Blue Shield’s complaint. While health systems are using machine learning to maximize billings, major insurance carriers have spent years building their own proprietary automated engines to aggressively deny coverage and truncate hospital stays. We have entered a bizarre era of algorithmic warfare where enterprise software applications are playing high-stakes poker with patient health outcomes.

"We are witnessing the birth of an automated financial ecosystem where machine learning models on the provider side write notes specifically designed to trick machine learning models on the payer side, while humans simply manage the API connections."

Consider the structural absurdity of this dynamic. On one side of the table, hospital systems implement sophisticated agentic workflows—similar in complexity to autonomous software frameworks like AX – Google’s Open Agentic Orchestrator—to automate clinical note synthesis, flag billing opportunities, and submit pristine claims. On the other side, insurance algorithms process those claims in milliseconds, flagging minor technicalities to trigger automated rejections. The hospital's AI then generates an automated appeal letter, which is subsequently parsed by the insurer's automated review system.

And that's the real story here. Millions of dollars are being spent on raw compute power, server bandwidth, and software licenses just to battle over existing pool dollars. Neither side is actually making the patient healthier during this transaction. The $942 million price tag isn't evidence that artificial intelligence is inherently flawed; it is proof that applying super-intelligent optimization software to an adversarial, highly bloated economic framework simply accelerates the rate at which money is drained out of the system.

How We Fix Healthcare AI Before It Breaks System Costs

If we want artificial intelligence to actually drive down the cost of medical care, we have to fundamentally change what we pay algorithms to do. As long as hospitals operate on a fee-for-service model, any AI tool introduced into the clinical workflow will be designed to maximize volume and documentation density. That is basic corporate strategy. If you want software to compress healthcare costs, you must tie its commercial success to value-based care models, where health systems are paid a flat fee to keep a population healthy.

Under a capitated, value-based model, the economic incentives invert completely. Suddenly, an algorithm that catches early signs of kidney failure before a patient requires hospitalization saves the system hundreds of thousands of dollars. An AI tool that helps a clinic manage diabetic care remotely reduces emergency room visits, preserving hospital margin. In that environment, "upcoding" yields zero financial benefit because billing density is no longer the primary revenue driver. Technology flourishes when financial incentives align with patient outcomes.

Let me be direct about this: regulators and enterprise buyers need