The FDA Just Asked the Internet How to Regulate AI Doctors
The FDA just asked the public how it should regulate generative AI medical devices, and the answer could reshape the entire health-tech industry. With AI device authorizations nearly tripling in two years, the clock is ticking on a regulatory framework that doesn't exist yet.
ChatGPT Walks Into a Hospital
Imagine your doctor pulls out a stethoscope, listens to your chest, and then asks ChatGPT what's wrong with you. That future isn't as far off as you think. And now the FDA wants to know: how should we handle this?
The FDA's Center for Devices and Radiological Health (CDRH) dropped a discussion paper asking the public a deceptively simple question. How should the government regulate medical devices powered by generative AI? Not the standard machine-learning tools that flag suspicious spots on an X-ray. We're talking about the newer, weirder breed: AI systems that generate new content, synthesize medical knowledge on the fly, and might one day talk directly to patients.
The comment period runs until October 19, 2026, filed under docket FDA-2026-N-7874. And if you care about the future of healthcare, this one matters.
Why This Isn't Just Another Bureaucratic Yawn
The FDA has been regulating AI-powered medical devices for years. That's not new. What is new is that generative AI breaks a lot of the old assumptions.
Traditional AI medical devices work like a really good spellchecker: they scan data, compare it to patterns, and flag things. A radiology AI looks at your brain scan and says, "Hey, this might be a stroke." It's narrow. It's predictable. Regulators can test it against known answers and sleep at night.
Generative AI is a different animal entirely. These systems can hallucinate (confidently make things up), drift in performance over time, and operate in ways that even their creators can't fully explain. Regulating them with the old playbook is like using a cookbook to fix your car; the tools just don't match the problem.
That's why the FDA is going back to basics and asking for help.
What the FDA Actually Wants to Know
The discussion paper isn't a new rule. It's not even a draft of a new rule. The agency was careful to say this is for discussion purposes only, and it doesn't change any existing policy. Think of it as the FDA raising its hand in class and saying, "We have some ideas, but we'd love feedback before we commit to anything."
Get tomorrow's biotech intelligence before your competitors.
Join thousands of biotech professionals who start their day with our free, daily briefing.
Those ideas center on four big areas:
1. Risk assessment. The FDA proposed a two-axis framework for sizing up how dangerous a generative AI device might be. The details are open for debate, but the core idea is giving regulators a structured way to say, "This one's low-risk, that one could hurt someone."
2. Premarket evaluation. Before a device hits the market, how should regulators test it? The paper floats a competency-based approach, combining non-clinical benchmarking (basically standardized tests for AI) with clinical confirmation in real-world settings.
3. Postmarket monitoring. Once a device is out in the wild, how do you keep tabs on it? This is especially tricky for generative AI, which can behave differently over time. The FDA wants monitoring proportional to risk: higher-risk devices get more scrutiny.
4. Foundation models and agentic AI. This is the spiciest section. Foundation models are the massive, general-purpose AI systems (think GPT-style architectures) that underpin many generative tools. Agentic AI refers to systems that can take actions on their own, not just answer questions. The FDA wants to know what special rules these might need.
The agency even floated an idea where AI model makers could voluntarily submit model cards or system cards, essentially nutrition labels for AI, to help device manufacturers document what's under the hood. Whether companies would actually do that without being forced is, let's say, an open question.
The Numbers Tell a Bigger Story
To understand why the FDA is moving now, look at the growth curve. The agency's list of authorized AI/ML-enabled medical devices has exploded. By the end of 2024, there were roughly 1,016 on the books. By end of 2025, that number jumped to 1,451. As of March 2026, it hit 1,524.
The pace is accelerating, too. In 2023, about 221 new AI devices were cleared. In 2024, that rose to 253. And in 2025, a whopping 295 were authorized. That's nearly triple the 2023 number in just two years.
Almost all of these (roughly 95% or more) came through the 510(k) pathway, which is the FDA's fastest route for devices that are similar to something already on the market. Only a tiny fraction, around 3–5%, used the De Novo pathway for truly novel devices. PMA approvals, the gold standard for high-risk devices, accounted for less than half a percent.
In other words, most AI medical devices are getting cleared by saying, "We're basically like that other thing you already approved." That works fine for traditional AI. For generative AI, which can behave in genuinely novel and unpredictable ways, the comparison gets a lot shakier.
The Elephant in the Exam Room
The real tension here isn't technical; it's political. There's an ongoing debate about whether the FDA even has enough legal authority to regulate generative AI devices properly, or whether Congress needs to step in with new powers. That debate is far from settled, and it's playing out against a backdrop of enormous commercial pressure.
Every major health-tech company is building generative AI into its products. Ambient AI scribes that write clinical notes. Diagnostic assistants that synthesize patient data. Systems that generate radiology reports. The market is moving fast, and the regulatory framework is, by design, still a blank page.
For biotech and health-tech companies, the practical takeaway is clear: expect the future to emphasize transparency, risk controls, and postmarket surveillance. If you're building a generative AI medical tool, start documenting everything now. The FDA may not have final rules yet, but the direction of travel is obvious.
The Bottom Line
The FDA just did something surprisingly humble for a federal agency: it admitted it doesn't have all the answers. This discussion paper is the starting gun for what will likely be years of rulemaking, but the fact that it's happening now, with generative AI still relatively early in its healthcare journey, is actually encouraging.
The comment window closes October 19. If you're a device maker, clinician, researcher, or even just a concerned citizen who'd rather not have an AI hallucinate your diagnosis, this is your chance to weigh in. The FDA is literally asking for your opinion.
How often does that happen?
Funding & Financings
5 min read
CSL Just Had Its Best Day in 25 Years. Here's What Lit the Fuse.
CSL posted a $2.6 billion net loss and its stock soared 18%, the best single-day move in 25 years. Behind the apparent contradiction lies a "reset year" ending, a massive buyback, and a plasma market where demand keeps outrunning supply.
