Artificial intelligence was supposed to make healthcare smarter, cheaper and easier. Instead, we may be watching it become the newest weapon in one of America’s oldest healthcare battles: who gets the money and who gets stuck with the bill.

 

A new analysis from the Blue Cross Blue Shield Association offers a remarkable example. Looking at its claims data, the association estimates that an increase in patients being coded as medically complex added $942 million in healthcare spending between 2023 and 2025. BCBSA says the increase happened as hospitals adopted AI-assisted coding tools that can scan medical records for diagnoses that might otherwise be missed.

 

That doesn’t necessarily mean the diagnoses are false. Medical coding determines how hospitals describe a patient’s condition and, in many cases, how much they are paid. A patient admitted for one problem may also have several secondary conditions buried in lab results and clinical notes. AI can search thousands of pieces of information far faster than a human coder and find conditions that qualify for additional reimbursement.

 

The problem is what happens when better documentation produces bigger bills without evidence of more treatment. BCBSA says it found a change in coding but no corresponding change in the care patients received. The organization argues that AI-assisted coding may therefore be helping move patients into more expensive billing categories without patients actually becoming sicker. Because BCBSA represents insurers that pay those bills, its conclusions should not be treated as those of a neutral observer, but the financial incentive it describes is real.

 

Hospitals are not the only ones bringing algorithms to the fight. Insurers are deploying AI to review claims, identify suspected overbilling and decide which claims deserve additional scrutiny. Reuters reported earlier this year that hospitals and insurers are increasingly using competing AI systems in their long-running battle over reimbursement. Hospitals use the technology to capture more revenue while insurers use it to control what they pay.

 

So much for eliminating administrative waste. We are automating it.

 

The hospital has an algorithm looking through your medical record for everything it can bill. The insurance company has another algorithm examining the resulting claim for everything it can challenge. When the insurer says no, the patient may increasingly turn to yet another AI system to understand the denial and figure out how to appeal it.

 

That last part matters because appeals are already difficult to navigate. KFF found that insurers selling plans through HealthCare.gov denied about 19% of in-network claims in 2024, yet consumers appealed fewer than 1% of those denials. The data covers only part of the insurance market and doesn’t tell us whether every denial was appropriate, but it shows just how rarely patients challenge them. KFF also notes that AI tools could help patients and healthcare providers prepare appeals.

 

That can be a genuinely useful application of AI. A chatbot can explain an insurance letter written in language most people rarely encounter, help organize records, identify information an insurer says is missing and help a patient draft an appeal.

 

But consider what we have created.

 

A hospital can afford sophisticated software to maximize reimbursement. An insurance company can afford sophisticated software to minimize unnecessary payments. Now a patient recovering from surgery, undergoing cancer treatment or simply trying to get a prescription covered may need their own AI assistant just to keep up.

 

People who are older, have limited computer skills, struggle with English, lack reliable internet access or are simply too sick to spend hours fighting an insurance company start this technological arms race at an obvious disadvantage. The more complicated the system becomes, the more valuable the tools needed to navigate it become.

 

And every layer costs money.

 

Hospitals pay for billing technology. Insurers pay for claims technology. Providers employ people to fight denials. Patients spend time appealing them. Those costs do not disappear; they become part of what Americans ultimately pay for healthcare.

 

AI isn’t necessarily the problem here. In many cases, it may be extremely useful. The problem is believing that smarter technology can fix a system whose incentives remain unchanged.

 

If one computer finds another diagnosis to bill, another computer finds a reason not to pay, and a sick person needs a third computer to figure out how to fight back, perhaps the breakthrough we need isn’t a better algorithm.

 

Perhaps we need a healthcare system that doesn’t require so many algorithms just to pay for healthcare.