LLMs for health advice get separate treatment
The disparity driven by AI is also reflected in how it's currently regulated. This is Part 1 of a three-post series looking at two examples that prove that point, and one that disproves it.
What started this
OpenAI's January 2026 report, "AI as a Healthcare Ally," revealed that more than 5% of all ChatGPT messages globally are health-related, and roughly 1 in 4 of its 800 million weekly users submits a health-related prompt every week — more than 40 million people doing so every single day. Over half of those users are checking or exploring symptoms.
We've all done it. And honestly, it's often impressively accurate, and blatantly diagnostic in everything but name.
Meanwhile, if you're the founder of a health assistant app, you're carefully calibrating every line of copy to avoid unsubstantiated medical claims. Why bother, when ChatGPT isn't? And how can a big-name LLM go seemingly unchallenged while doing, at scale, exactly what your product is regulated for doing at a fraction of the volume?
The answer is intended use
A symptom checker or AI health assistant is explicitly built and marketed for health screening. That intended use — which also drives your marketing copy, whether you realise it or not — is what determines whether something is or isn't a medical device, under both FDA and EU MDR frameworks.
ChatGPT and its peers are, officially, general-purpose LLMs. Under EU MDR guidance, software can escape the medical device definition specifically because its intended use is "too generic" — a model trained for broad-purpose conversation, not fine-tuned or marketed toward a specific medical purpose, sits outside the framework by design. The same logic runs through FDA's approach: it's not the model's technical capability that triggers regulation, it's what the manufacturer says the product is for.
That's why ChatGPT and its peers cannot legally be advertised for health purposes, even as OpenAI's own data shows the capability — and the demand — is clearly there. OpenAI has been careful in how it frames its January 2026 launch of ChatGPT Health, positioning it as designed to "support, not replace" medical care, precisely to stay on the right side of that line.
The enforcement gap is real, not just theoretical caution
Unofficially, as long as a general-purpose LLM isn't "misbranded" — advertised for diagnostic use — enforcement relies on demonstrating a genuine misuse risk. That's a harder bar to clear than a simple intended-use check, and clearing it takes time regulators don't always have relative to how fast these products are shipping.
Legal scholars have started making the case directly: some argue health AI chatbots already meet the legal definition of a medical device given how they're actually used, and that regulators simply haven't caught up to treating them accordingly. Meanwhile, reporting through 2026 has flagged that the FDA appears to be standing back from active oversight of medical chatbots specifically, leaving individual states to step in — a Pennsylvania lawsuit against a chatbot provider is one visible example of that gap being filled by litigation rather than regulation.
The open questions nobody has settled
Should intended use still be the deciding factor, when the same underlying model can be marketed two completely different ways with two completely different regulatory outcomes? Should general-purpose LLMs be regulated regardless of stated intent, given how they're actually used at scale? Or should providers be required to restrict diagnostic-style output altogether, rather than relying on disclaimers?
Different regulators are landing in different places, and the topic is heavily debated — and heavily lobbied — across jurisdictions, evolving faster than most guidance documents can keep pace with.
Different players, different bets
Companies are hedging this uncertainty in visibly different ways. Some lean hard into wellness framing and disclaimers — Ōura and WHOOP are the clearest examples, sitting deliberately on the general-wellness side of the FDA's own wellness guidance, a line I've written about before given how much latitude that guidance currently affords non-diagnostic wearables. Some build a dedicated, more clinically-postured offering, as OpenAI has done with ChatGPT Health. And some appear to be simply waiting it out, for as long as the current enforcement gap allows.
Beneath the apparent stillness, both regulatory frameworks and business strategies are shifting quickly. Neither is settled, and betting on today's gap staying open is itself a strategic choice, not a neutral one.
Will trust end up doing regulation's job?
After OpenAI's recent trust-related scrutiny — including its own disclosures about how it handles the small percentage of users showing signs of mental health crisis in conversation — I find myself wondering whether consumer trust ends up doing more of the practical work here than formal regulation does, at least in the near term. If that plays out, it could genuinely benefit the players who treat medical-grade rigor as a differentiator now, rather than a compliance cost to defer.
Meanwhile, we can help you find the right balance between ambition and defensible claims — reach out if you're navigating this line yourself.
References
OpenAI, AI as a Healthcare Ally: How Americans Are Navigating the System With ChatGPT, January 2026
Petrie-Flom Center, Harvard Law School, Health AI Chatbots are Legally Medical Devices; It's Time the FDA Started Treating Them Like It, May 2026
Akin Gump, States Confront Oversight of Medical Chatbots While FDA Stands Down, May 2026
Giulia Paggiola, FDA's new guidance on general wellness, Edge Compliance blog — edgecompliance.co/blog/fdas-new-guidance-on-general-wellness
Methodology note: This article is based on Part 1 of my original three-post LinkedIn series (link), reflecting my professional experience and perspectives on how AI regulation is currently applied to general-purpose LLMs versus purpose-built health tools. AI assisted in elaborating the topic into a broader article by integrating background research, fact-checking of the underlying statistics, and additional regulatory references. All analysis and regulatory perspectives are my own, and all content has been reviewed by me for accuracy.
