Who Is Responsible for AI Coding and Billing Errors

The adoption of artificial intelligence in healthcare is reshaping how medical professionals handle administrative tasks, but the shift introduces significant risks for coding and billing accuracy. Ambient AI scribes are increasingly used to transcribe patient interactions in real time, yet these systems can omit, misinterpret, or fabricate clinical details. This creates a liability where practices might submit claims based on inaccurate information, potentially leading to upcoding and false claims against government insurers.

Modern “ambient AI” systems use recording devices to capture patient conversations automatically rather than relying on clinicians to type notes or dictate them later. The AI then generates a transcript, and a coder reviews this text to select the appropriate medical codes. These generated encounter notes are often difficult to interpret due to missing time elements, a lack of specificity, and insufficient details about medical decision-making. One study from 2025 found a 1% to 3% overall error rate in modern ambient AI scribes, with authors noting that not only was information misinterpreted or omitted, but in some cases, the system completely fabricated fictitious content.

In one scenario, a family practice physician sees an established patient for a routine annual physical. The physician asks the patient about monitoring blood pressure, and the visit concludes without any medication changes. The AI scribe misinterprets this question and creates documentation indicating the patient was seen for high blood pressure. The AI coding system then instantly converts the visit from a preventive encounter to a problem visit. The insurer is billed for code 99214, which pays about $135, when the claim should have used code 99396 for a preventive visit, paying only about $128. This small difference results in the practice upcoding the visit and receiving more payment than it was entitled to.

For clinicians, the immediate benefit of these tools is a reduction in documentation time. AI documentation is believed to save up to 30% in time for clinicians, offering a potential efficiency gain that appeals to practices looking to devote more time to patient care. However, this speed comes with a trade-off. If an AI system is selecting diagnosis and procedure codes without human oversight, the potential for compounded miscoding issues rises sharply. This is particularly risky when no human reviewer is involved at all, as the AI scribe creates the transcription and an AI tool selects the corresponding codes.

Liability Under the Federal False Claims Act

Regardless of whether documentation is done manually or through AI, the physician remains ultimately responsible for the diagnosis and procedure codes billed on claims. The Federal False Claims Act (FCA) states that submitting false or misleading claims to government insurers exposes practices to penalties and liabilities. Even if improper coding isn’t intentional, reckless disregard can be enough for the government to impose penalties. The US Department of Health and Human Services (HHS) Office of Inspector General lists improper claims as those linked to services that were medically unnecessary, not actually performed, performed by excluded staff, or included in a global package when they shouldn’t be.

Consider a psychiatrist using ambient AI to document a 10-minute medication refill visit. The physician asks the patient about therapy attendance, and the patient replies that they attend 60 minutes of therapy weekly. The ambient AI produces documentation indicating the psychiatrist performed a 60-minute therapy session and a medication refill. The practice then submits claims for both 99212 and 838, paying about $138 more than the visit warranted. Since psychotherapy wasn’t actually performed, this specific claim could trigger a False Claims Act violation suit, highlighting how AI errors can lead to financial and legal repercussions.

Practices are encouraged to implement formal governance policies to mitigate these risks. A June 2026 report by the health law firm Baker Donelson indicated that 63% of healthcare organizations had no AI governance policies in place, even though many were already using the technology. To protect against errors, it is essential to review all AI-generated documentation and billing recommendations before claim submission. Your practice should have an internal audit process where documentation is periodically spot-reviewed by someone else at the office to ensure the codes match what was actually done on the date of service. Working with a healthcare attorney to create a formal AI governance policy can help delineate exactly how and when AI is used to maintain proper checks and balances.