FDA clears first medical software using a patient-facing large language model
Digital-health company UpDoc disclosed that it has received FDA clearance for a diabetes-management app that patients interact with directly through a large language model, using voice and text. STAT reports it as the first US clearance for medical software built on a patient-facing LLM — meaning the AI speaks to patients rather than working behind the scenes to support clinicians.
The app operates within a treatment plan set by a patient’s doctor, and the FDA placed it in the same regulatory category as drug-dose calculators that recommend insulin amounts based on inputs such as blood glucose. That framing matters: it treats the tool as a bounded clinical calculator rather than an open-ended chatbot.
The unresolved question is how much the generative model actually decides. UpDoc’s chief executive would not say whether the AI makes treatment decisions or merely relays doctor-defined guidance — a distinction that goes to the heart of how regulators will judge the next wave of patient-facing AI, where the interface and the decision-maker can blur together.