Artificial intelligence was supposed to make healthcare faster, smarter and potentially cheaper.

Instead, a new analysis from the Blue Cross Blue Shield Association is raising an uncomfortable possibility: some of the first financial effects of hospital AI may be showing up not as lower medical bills, but as higher ones.

BCBSA says changes in hospital coding associated with the growing use of AI tools contributed an estimated $942 million in additional healthcare spending for its member insurers between 2023 and 2025. The organization says more patients were increasingly classified as having medically complex cases, even though its claims analysis found no corresponding change in the intensity of care delivered.

That finding does not establish that every additional diagnosis was inappropriate or that AI itself directly caused every dollar of the increase.

But it opens a much bigger question about how artificial intelligence is changing the economics of American healthcare.

And consumers could eventually feel those changes through premiums and out-of-pocket costs.

BCBSA analyzed de-identified claims data from its member health plans, which collectively cover more than 100 million people, or roughly one in three Americans. The association looked at inpatient claims from the beginning of 2023 through the end of 2025 and found that the proportion of hospital cases classified as medically complex increased from roughly 37% to 40%.

The association estimates that the resulting increase in reimbursement cost approximately $942 million for Blue Cross plans over two years.

About $653 million of that increase was linked to secondary diagnoses — conditions beyond the primary reason for a hospital stay — that moved claims into higher-reimbursement categories. BCBSA says more than 55,000 additional cases entered those higher-complexity categories relative to its 2023 baseline.

That is where the AI connection enters.

Hospitals increasingly use software that can examine electronic health records, physician notes, laboratory results and other data to identify diagnoses that can be included in a claim.

A doctor may treat a patient for one primary condition.

An AI-assisted system can scan the patient's medical record and identify additional conditions such as anemia or low sodium.

Under hospital reimbursement systems, those secondary diagnoses can sometimes move a patient's stay into a more expensive diagnosis-related group.

The technology is extremely good at finding information.

The debate is over what that information means financially.

This is the centerpiece of the Blue Cross analysis.

BCBSA says the rise in coding complexity did not come with an equivalent increase in several measures of treatment intensity.

For example, the association examined major bowel procedures and found that cases classified at the highest level of complexity increased sharply between early 2023 and late 2025. It also examined indicators such as blood transfusions, intensive-care utilization, reoperations and length of stay.

The insurer group's argument is that if patients were genuinely becoming significantly sicker, more intensive treatment would generally be expected to accompany those diagnoses.

Instead, BCBSA says it observed a widening gap between what the billing codes suggested and what the claims showed hospitals were actually doing.

That is why the association describes the additional spending as a concern rather than simply a technological success.

The healthcare industry has pushed back against the suggestion that higher coding complexity automatically means hospitals are improperly billing insurers.

The American Hospital Association argues that hospital patients have become older and medically more complex over time, while many less-complex procedures have shifted from inpatient settings to outpatient care. That can make the remaining inpatient population appear sicker even if individual hospitals have not suddenly changed the way they treat patients.

This matters because BCBSA's analysis is based primarily on claims data rather than complete clinical charts.

The insurer association itself acknowledges that its analysis does not prove AI caused every individual increase in coding.

It also does not identify specific hospitals or AI vendors responsible for the changes.

That leaves room for multiple explanations.

Hospitals may genuinely be documenting conditions that previously went unrecorded.

AI may be making that documentation faster and more complete.

Hospitals may also have stronger financial incentives to capture every legitimate diagnosis.

Or some combination of those factors could be occurring simultaneously.

The claims data show the financial result.

They do not, by themselves, settle the argument over why every individual diagnosis appeared.

The reason ordinary Americans should care is the structure of the insurance system.

When insurers pay more for medical claims, those expenses do not simply disappear.

Over time, higher medical spending can contribute to pressure for higher premiums, deductibles, copayments or other out-of-pocket expenses.

BCBSA says that additional costs can ultimately affect families, employers and taxpayers through the healthcare financing system.

That comes at an already difficult moment.

The median proposed increase in individual-market health insurance premiums for 2027 is around 15%, according to analysis cited by Yahoo Finance, following a median increase of about 20% for 2026.

Healthcare consumers are therefore confronting multiple cost pressures at once.

Prescription prices remain a concern.

Hospital services remain expensive.

Labor and administrative costs continue to affect providers.

And insurers are trying to manage rising claims.

Now AI-assisted coding has entered that equation.

Even if its contribution is relatively small compared with the total U.S. healthcare bill, a $942 million increase across Blue Cross plans is large enough to attract serious attention.

There is an important twist.

Insurers are also using artificial intelligence.

BCBSA's report comes amid a wider expansion of AI tools throughout the insurance industry, where companies use automated systems to process claims, identify inconsistencies and evaluate whether requested care meets certain criteria.

The federal government is experimenting with AI-assisted approaches as well.

CMS's Wasteful and Inappropriate Service Reduction, or WISeR, model uses technology in a prior-authorization system for certain procedures and services. Yahoo Finance reported that critics have raised concerns about delays and denials associated with the approach.

That creates a remarkable new dynamic.

Hospitals can use AI to find additional diagnoses.

Insurers can use AI to challenge claims.

The result is a system in which algorithms increasingly sit on both sides of the financial transaction.

One side is searching the patient's record for information that could affect reimbursement.

The other side is searching the claim for information that could affect whether reimbursement should occur.

Humans remain involved, but the software is becoming far more powerful.

There is an important distinction between an AI system identifying a legitimate diagnosis and an AI system being used to justify an inappropriate claim.

Artificial intelligence can clearly improve healthcare administration.

It can reduce paperwork.

It can help clinicians document encounters.

It can surface information that a human might miss.

And in some settings, better documentation can result in more accurate reimbursement.

The concern arises when reimbursement incentives become more powerful than the relationship between documentation and actual care.

BCBSA says its analysis identified precisely that disconnect.

The association found that the share of patients classified as medically complex increased substantially, but it did not find evidence of a similar increase in treatment intensity across the measures it examined.

Hospitals, however, say AI can simply make documentation more accurate.

That disagreement is likely to become increasingly important as adoption spreads.

There is another issue hiding underneath the controversy.

AI companies often sell their technology on the promise of efficiency.

In healthcare, efficiency theoretically means reducing administrative workloads, speeding documentation and allowing medical professionals to spend more time with patients.

But if the technology simultaneously changes how providers code and bill, it can create a second economic effect.

Better information can mean more recognized diagnoses.

More recognized diagnoses can mean higher reimbursement.

Higher reimbursement can mean greater spending.

So the question is not simply whether AI works.

It is whether the financial incentives surrounding the technology are aligned with the goal of reducing unnecessary costs.

That is much harder.

The Blue Cross findings are likely to attract attention because healthcare represents one of the largest areas of economic activity in the United States.

If AI adoption leads to large-scale changes in medical coding, even relatively small effects per hospital stay can accumulate into billions of dollars.

That means regulators may eventually need clearer rules around AI-assisted documentation and coding.

Insurers will want to know how diagnoses are generated.

Hospitals will want protection for legitimate reimbursement.

Patients will want confidence that their medical records are not being manipulated by automated systems for financial purposes.

And AI vendors will have to demonstrate that their systems improve accuracy without creating unexpected incentives.

The most important question may eventually be much simpler than the technical debate.

Did AI improve the patient's outcome?

Did it reduce the doctor's workload?

Did it prevent a mistake?

Did it shorten the hospital stay?

Did it reduce costs?

Or did it simply find more diagnoses that could be attached to a claim?

Those questions cannot be answered by billing data alone.

They require clinical evidence.

That is why the BCBSA study should be treated as an important warning signal, not as a final verdict on healthcare AI.

The $942 million figure is substantial.

The rise in coding complexity is measurable.

The lack of a corresponding increase in treatment across the indicators BCBSA studied is notable.

But the underlying cause remains contested.

Artificial intelligence is only becoming more deeply embedded in America's healthcare system.

Hospitals are using it for documentation and coding.

Doctors are using it for clinical support.

Insurers are using it for claims processing.

Government programs are experimenting with automated decision-support tools.

The technology is becoming unavoidable.

That makes the incentives surrounding it increasingly important.

AI could eventually make healthcare more efficient and less expensive.

But the latest Blue Cross analysis shows another possibility: technological efficiency at one point in the healthcare chain can create new costs somewhere else.

For patients already facing rising premiums, that distinction matters.

The future healthcare bill may not simply reflect how much medicine America consumes.

It may also reflect how intelligently — and how financially — artificial intelligence interprets every line of the medical record.