AI is reshaping how US hospitals bill and collect medical debt

Debt collection gets a new tool

Medical debt is likely to grow among both insured and uninsured Americans, said April Kuehnhoff, a senior attorney at the National Center for Consumer Law. Some people lose coverage because they cannot afford it or meet stricter requirements; others end up with plans that cost less but leave them with higher deductibles and copays.

The pressures are expected to worsen. Donald Trump’s HR1, known as the One Big Beautiful Bill Act, cuts federal health-care funding by about $1 trillion. Combined with Republicans’ decision not to extend subsidies for Obamacare plans, the measures are projected to leave about 17 million people uninsured by 2034.

More uninsured patients usually means more uncompensated care: bills for treatment that hospitals and doctors provide to people who lack adequate coverage and cannot pay. The Urban Institute estimates that HR1 will add $278 billion in uncompensated care by 2034. Hospitals short on funds have a reason to pursue unpaid bills more aggressively—and to turn to technology companies to do it.

Cedar, an AI billing platform backed by investors including Andreessen Horowitz and JP Morgan, says it uses hospital and physician data to predict patients’ willingness to pay and select discounts. Its chatbot, Kora, handles billing questions, with the aim of reducing calls to billing departments and keeping staffing costs down.

The company also checks whether patients have coverage, including Medicaid, using questionnaires—a longstanding practice in health care. Cedar strategy chief Dugan Wincie said the company has worked with uninsured people and those eligible for Medicaid for several years. It aims to treat a single mother with $10,000 in debt differently from a former executive who owes a $50 copay, and to make options such as interest-free payment plans more visible to patients who have trouble paying or rarely use digital services.

Debt buyers are using AI too, to estimate how much to offer for portfolios of medical debt. The broader promise of AI in health care has often been framed as curing disease; Anthropic CEO Dario Amodei, among others, has promised that AI could help defeat all diseases within a decade. Researchers are skeptical. JP Morgan financial analysts, by contrast, have pointed to more modest early applications in coding, billing and collections.

The bill can grow before it reaches collections

AI-assisted medical coding has drawn scrutiny for a different reason: it can increase the amount billed. In spring 2026, the Blue Cross Blue Shield Association (BCBS), which represents regional Blue Cross insurers, reported that women were increasingly being billed after severe blood loss during childbirth even though the rate of blood transfusions had not changed.

BCBS described a process in which AI scribes turn conversations between patients and clinicians into structured records. An AI program then uses those records and the patient’s electronic health record to suggest more detailed service codes, which can raise the bill. In the researchers’ account, the technology was increasing hospital charges.

For companies selling AI scribes, more precise coding is a selling point. One company backed by Andreessen Horowitz told investors that its recording software brought clinics an additional $13,000 in annual revenue per doctor through more accurate coding. A second BCBS report, published in September, said AI-assisted coding had added $1 billion to health-care costs in a year.

The American Hospital Association disputed what the analyses establish, saying they “raise questions about coding trends, but do not prove that hospitals are coding patient data incorrectly or that AI tools lead to inappropriate spending.”

That dispute sits inside a longer fight between insurers and health-care providers over who is responsible for rising costs. Insurers’ own AI use has also drawn criticism: opponents say opaque algorithms can delay or deny care. The Department of Health and Human Services under Robert F. Kennedy Jr. recently launched a program in which Medicare will use AI for prior authorization of treatment for some conditions, an approach already used by private insurers.

Personalized relief, personalized pressure

I think the most revealing feature of Cedar’s pitch is not that it automates billing, but that it uses patient data to tailor what a person is asked to pay. A discount could make a bill manageable; the same prediction of willingness to pay could also help a collector decide how much to pursue. The announcement leaves unclear how those decisions are bounded, or how patients can challenge them.

AI is entering health-care finance at both ends: it can help produce a larger bill through coding, then help determine how to collect it. As more patients face costs they cannot easily absorb, the distinction between making payment simpler and making collection more effective will matter as much as the software itself.