The first class-action lawsuit against a health system over AI scribe consent was filed before most clinical leadership teams knew what an ambient documentation tool was. That is the part worth sitting with.
The technology rolled out fast. In 2024 and 2025, health systems signed contracts, completed pilots, and expanded to hundreds of clinicians with remarkable speed. Patient consent documentation followed at a different pace, if it followed at all.
Now three health systems are in federal court. California, Washington, and eleven other states have all-party consent laws: before a recording device, digital or human, captures a clinical encounter, every person in the room must know it is happening and agree. The legal standard was not ambiguous. The implementation timeline was.
I want to be precise here, because the noise around this story is already trending toward "AI is dangerous" rather than "this is what happens when you deploy faster than you govern."
The scribe model did not make a decision. It recorded what it was configured to record. The failure was upstream: the consent workflow was written after the software was already running, legal was looped in late, and the patient was asked at the wrong moment, or not at all.
This is a pattern I have watched play out in clinical settings for twenty years, with every class of new technology. The pilot starts under a champion who moves fast. The compliance team gets a one-page brief the week before expansion. The intake staff gets a training module. The patient gets a new line on a form they are handed while putting on a gown.
"When ambient AI is running during that encounter and the patient was not told, the form is not enough. The conversation had to happen before the session began."
At core, these cases allege that patients were recorded during clinical visits without their knowledge and without the consent their state law required. The AI companies are named alongside the health systems. The damages theory is still forming, but the discovery process alone will be expensive, time-consuming, and reputationally uncomfortable.
The question clinical leadership should be asking is not "could we lose this case." It is "what did our consent workflow look like at every site, on every platform, at every point in the rollout."
If you cannot answer that in 48 hours, you have a governance problem.
Health systems that have not yet faced a challenge of this kind have a window. The consent framework for ambient AI documentation should include:
That is not a technology problem. It is a workflow design problem. The solution is a consent script, a training protocol, and an intake process that puts the patient ahead of the efficiency metric.
The AI scribe vendors will tell you their product is compliant. Compliance with what the vendor certifies and compliance with your state's consent statute are not the same sentence.
Health system boards that approved these AI implementations without asking "what is the consent process" have a lesson here that will cost somebody. If the clinical teams reporting up did not know the regulatory framework, that is a training gap. If they knew and did not escalate, that is a governance gap.
Either way, the answer is not to slow down AI adoption. The answer is to build the governance architecture the technology deserves.
That work is not glamorous. It does not generate press releases. But it is what keeps a class-action docket from becoming the headline.
Surgery-trained, currently practicing internal medicine. Advisory practice focused on clinical AI governance, vendor evaluation, and implementation strategy for health systems and health-tech companies.
Weekly clinical reality on AI, governance, and what actually works inside health systems.