Artificial intelligence (AI) tools are proliferating fast in health care, but the legal framework around them is still catching up.

In this episode, Dan Silverboard, J.D., a health care attorney at Holland & Knight, joins Medical Economics Managing Editor Todd Shryock to explain how AI is currently being regulated — by states primarily, and by the FDA only indirectly — and where the biggest liability gaps exist for physicians and practices. He walks through what happens legally when an AI-generated recommendation contributes to patient harm, why the responsibility almost always lands on the provider, and why there is no get-out-of-jail-free card when an AI tool generates a higher billing code than what was actually performed.

Music Credits:
Sky Drifter by Cephas - stock.adobe.com
A Textbook Example by Skip Peck - stock.adobe.com

Editor's note: Episode timestamps and transcript produced using AI tools.

0:00 – 0:25 | Sponsor message Copic medical liability insurance.

0:25 – 0:51 | Cold open Silverboard delivers the episode's central warning: periodic auditing of AI-generated billing documentation is non-negotiable, and there is no get-out-of-jail-free card when an AI tool recommends a higher code than what was performed.

0:51 – 1:53 | Introduction Austin Littrell introduces the episode and previews the conversation with Silverboard.

1:53 – 3:09 | How AI in health care is currently being regulated Silverboard explains that states are the primary regulators, treating AI as a technology that supports clinical decision-making rather than a medical device. The FDA regulates AI only indirectly, based on whether it's incorporated into a regulated medical device.

3:09 – 4:57 | The two biggest liability risks for physicians using AI Silverboard identifies the core risks: 85% of health care AI investment is going to startups without proven compliance track records, and providers who blindly sign off on AI recommendations — clinical or documentation-based — without verifying accuracy are taking on serious legal exposure.

4:57 – 6:45 | Who is liable when AI contributes to patient harm Silverboard explains that legally, the provider must sign off on any AI recommendation, making them the primary responsible party. Technology vendors can face liability if their product is found to be wholly deficient — trained on biased or false data, for example — but broad liability disclaimers in vendor contracts make that a high bar.

6:45 – 7:28 | Should physicians document AI use in the medical record Silverboard says yes — physicians should document whether AI was used, whether they followed its recommendations and, if they deviated from them, why. Several states, including North Carolina, have already passed legislation or board guidance requiring exactly this.

7:28 – 8:37 | Compliance and billing risks from administrative AI tools Silverboard is direct: providers attest to the accuracy of their claims, and that responsibility doesn't transfer to an AI tool. Up-coding, down-coding and unbundling errors generated by AI are still the provider's problem. Periodic auditing and monitoring of all billing documentation — AI-generated or not — is essential.

8:37 – 9:27 | P2 Management Minute Keith Reynolds shares practice management tips and invites listeners to submit their own workflow ideas.

9:27 – 10:36 | What to demand in an AI vendor contract Silverboard outlines the must-haves: robust HIPAA compliance representations and warranties, ongoing validation and bias testing with reporting obligations, and a data governance plan confirming the AI system and its training data are free from bias or untrustworthy sources.

10:36 – 12:30 | Privacy complications when AI learns from patient data Silverboard explains a key HIPAA limitation: vendors can only train on protected health information for the benefit of the contracting provider — not to improve their own product. De-identified data is simpler, but practices still need contract provisions prohibiting re-identification, which is an increasingly realistic risk as AI becomes more powerful.

12:30 – 13:24 | Legal concerns around ambient AI and automated note generation Silverboard says the core risk is providers relying too heavily on ambient AI without verifying that the record accurately reflects the encounter. Texas has already codified this as a statutory requirement for all providers using AI to record patient encounters.

13:24 – 15:13 | Three questions practices should ask before deploying AI Silverboard's framework: first, vet the vendor thoroughly for HIPAA compliance and a proven track record; second, understand your patient population's comfort level with AI, which should shape how and where you deploy it; and third, decide how you will disclose AI use to patients — regardless of whether your state requires it.

15:13 – 16:16 | Where AI-related litigation is heading Silverboard says if HHS projections hold, AI could actually reduce adverse events and litigation over time. But one area he expects to keep growing: lawsuits challenging health insurers' use of AI to deny or down-code claims and prior authorization requests.

16:16 – 17:15 | Closing thoughts and outro Silverboard closes with a note of optimism — AI holds great promise — paired with a practical bottom line: vet your vendors, monitor your billing, and build compliance checks into your program now. Littrell thanks listeners and reminds the audience to subscribe and visit MedicalEconomics.com and PhysiciansPractice.com.

S1 Ep142: The legal risks of AI in your practice, with Dan Silverboard, J.D., of Holland & Knight

Off the Chart: A Business of Medicine Podcast