Three years on, the question is no longer whether AI belongs in the visit, but who answers for it
When we first wrote about AI in telemedicine in 2023, most of it was still a demonstration. A chatbot could draft a letter. An algorithm could flag a nodule on a chest film. The practical question for a telehealth practice was whether to try any of it. That question has been answered. Ambient scribes now listen to the video visit and write the note. Decision-support tools suggest a differential before the clinician has finished the history. Patient-portal messages are drafted by a model and sent after a glance. The FDA's list of AI-enabled medical devices passed 1,200 in 2025, most of them in imaging.
What has not changed is where the liability lands. When the note is wrong, the scan is misread or the portal reply gives bad advice, the claim is a malpractice claim against the clinician and the practice. The vendor's contract will say so. This article is about where AI sits in a telemedicine practice today, what the law has started to say about it, and what the malpractice carrier expects to see on the application.
Where AI is in the visit now
- Documentation. Ambient scribes record the consultation and produce the note. They save time and they make a particular kind of error: a confident, well-written note that says something the patient did not say. A note signed without being read is the clinician's note.
- Decision support. Tools that suggest diagnoses, flag drug interactions or propose an order set. Useful, and bounded by what they were trained on. Rare presentations and unusual patients are exactly where they fail quietly.
- Imaging. Algorithms that read X-rays, dermatology photos and retinal images are the largest group on the FDA's list. In a telemedicine setting they often arrive before a specialist does, which tempts a practice to treat the algorithm as the specialist.
- Patient communication. Models that draft portal replies, triage messages and after-visit instructions. This is the use that state legislatures reached first.
- Remote-assisted procedures. Robot-assisted surgery with a remote component exists, but it is a hospital question rather than a telehealth one, and the liability sits with the facility and the manufacturer as much as the surgeon.
What the law now says
The 2023 article said the legal framework was coming. Some of it has arrived, state by state.
- California, since January 1, 2025. Under AB 3030, a facility or practice that uses generative AI to write a patient communication about clinical information has to say so in the message and tell the patient how to reach a human. The duty falls away when a licensed clinician reads and reviews the message before it goes out, which is the point: review it, or label it.
- Illinois, since August 2025. The Wellness and Oversight for Psychological Resources Act bars AI from providing therapy or making therapeutic decisions. It may handle scheduling, billing and administrative support. The state licensing department can fine up to $10,000 per violation. Telepsychiatry and teletherapy platforms serving Illinois patients have to show a licensed clinician in the decision.
- Federal nondiscrimination. Under Section 1557 of the Affordable Care Act, covered providers must make reasonable efforts to identify and reduce discrimination arising from the patient-care decision-support tools they use. A practice is responsible for checking that the tool does not treat some patients worse, not just for buying it.
- The FDA's position on the devices it clears is that the clinician stays in the loop and remains responsible for validating what the algorithm produces.
None of this creates a new standard of care for AI. It does something simpler: it puts in writing that the human is responsible, which is also what the courts have assumed all along.
Data, privacy and bias
Two risks from 2023 have only grown. The first is data. A telemedicine practice using cloud AI tools is sending recordings, images and records to a third party, and a breach at the vendor is a breach the practice has to notify and answer for. The business-associate agreement and the practice's own cyber liability cover carry that. The second is bias. A tool trained on one population can underperform on another, and under Section 1557 that is now a compliance problem as well as a clinical one. Ask the vendor what the tool was validated on, and keep the answer.
What the malpractice carrier wants to see
Telemedicine malpractice underwriters have settled into a clear position on AI, and it matches the law:
- A clinician reviews and signs off every decision. Tools that draft notes, triage messages or suggest a differential are insurable when a licensed clinician reviews the output and the workflow says so. A service that lets a model or a store-and-forward queue reach the patient unreviewed is the exposure underwriters refuse, or price heavily.
- The workflow is described on the application as it actually runs. Which tools, for which tasks, who reviews, how it is documented. A gap between the application and the practice is the first thing a defence lawyer and a claims adjuster will find.
- The vendor contracts are read. Most AI vendors disclaim clinical liability and cap their exposure at the fees paid. The practice should know that before an incident, not after.
- Cyber liability sits alongside malpractice. Malpractice responds to the harmed patient; cyber responds to the breach, the ransom and the downtime. A practice running on cloud AI tools needs both, and should check which policy excludes what.
For a telehealth group, the application question on AI is worth answering carefully rather than briefly. Our telemedicine malpractice insurance page sets out current pricing and the factors carriers rate, including AI-assisted diagnosis.
For agents and brokers
The 2023 article was written for commercial agents, and the advice stands: stay current on the state laws, read the vendor contracts your clients have signed, and talk to the underwriters, who are seeing the claims before anyone else is. Carrier briefings on telemedicine and AI are part of the regular calls at Insurance Insiders.
