US medical debt is growing, and AI will increasingly be the one collecting it

According to KFF estimates, about 100 million Americans already carry debt for medical services. The Guardian reports that Republican-initiated cuts to health programs take effect this winter, and the volume of such debt is expected to grow. At the same time, the mechanics of billing itself are changing. AI listens to the appointment, codes the visit and generates billing messages, while AI agents exchange messages with patients in payment portals. A debt is now created and serviced inside a single automated chain.

One chain instead of four departments

It used to work like this: the doctor wrote the note, a coder assigned the codes, the billing department issued the bill, and collectors handled the debt. Now every link gets an AI layer. The Guardian's reporting shows the following:

• Appointment. Assistants record the conversation between doctor and patient and draft the clinical note.
• Coding. Algorithms convert the note into billing codes, which the bill is built on.
• Messaging. Systems automatically send notifications about charges and payments.
• Agents. AI bots write to patients in the payment portal, remind them about overdue balances and answer questions.

The available excerpt of the article names no vendors and gives no figures on deployment scale or metrics. But the logic is plain. The note determines the code, the code determines the amount, and the amount determines what the agent writes and to whom. In effect, the neural network at the first step sets everything that follows.

Why now

The program cuts mean more people will end up uninsured or with worse coverage. The share of bills that patients must pay out of pocket will rise. These remainders are the ones most often turned into overdue debt.

Automation changes the economics of collection. A single human phone call costs money; a single message from an agent costs almost nothing. When the cost of a contact falls, the number of contacts rises: reminders arrive more often, at a time convenient for the debtor, tailored to their history in the portal. That is my observation, not a conclusion of the article, but it is how the industry's logic works.

The regulatory backdrop remains uncertain. A court vacated the federal CFPB rule that was meant to remove medical debt from credit reports in 2025. It is still unclear how strictly AI communication with debtors will be restricted.

Where the system can break

The first risk zone is errors that propagate down the chain. If the assistant mishears or missummarizes the appointment, and the algorithm codes the visit as more expensive than it was, the patient will receive an inflated bill. Neither the doctor nor the person in billing may notice.

The second is pressure. An agent that knows when a patient logs into the portal can write at exactly those minutes. In such systems, the line between a reminder and pushing is set by configuration, not by human tact.

The third is disputed cases. A bot is good at reminding someone about a payment but worse at working out why a bill is wrong. If escalation to a human is buried deep, the patient is left alone with the machine.

What patients and companies should do

Patients should request an itemized bill with codes and check it against what actually happened at the appointment. It is better to dispute in writing and before the debt is handed to collectors. Nonprofit hospitals in the US are required to have financial assistance programs; ask about them before paying, not after.

Clinics and vendors should keep logs: which note produced which code, and which message went to the patient. They need sampled audits of coding, limits on the frequency and timing of agent contacts, and a "talk to a human" button on the first screen. Without these, savings on billing quickly turn into legal claims and lost trust.