Health AI Chatbots are Legally Medical Devices; It’s Time the FDA Started Treating Them Like It

In January 2026, OpenAI launched ChatGPT Health. OpenAI promises ChatGPT Health will help users clarify medical concepts, understand test results, prepare for appointments, and explore healthcare and wellness strategies — all while leveraging user-uploaded health records. Yet the product name itself is misleading: The marketing of the “health” chatbot asserts the product is not a diagnostic or symptom-screening tool. OpenAI plausibly sought to circumnavigate the definition of a medical device. However, mere manufacturer claims of intended use are not enough to escape regulation by the Food and Drug Administration (FDA).

Indeed, the public may be alarmed to find no mention of safeguards against misuse (such as using ChatGPT Health to diagnose a mystery ailment) featured in the product’s advertisements, despite OpenAI acknowledging that users routinely ask the original model health-related questions.

If OpenAI wants to play doctor, it has to play by the rules.

Regulating Health AI Chatbots

Broadly speaking, AI chatbots engage in simulated conversation using humanlike language. Despite the widescale integration of chatbots into healthcare, the FDA leaves them unregulated. Most chatbots are directly tied to a regulated payer or provider — such as the recently launched Amazon Health AI platform. Other chatbots do not connect patients with care, but nonetheless maintain some standard of regulation. ChatGPT Health is wholly separate: an unregulated chatbot without any legitimate commitment to safeguard its users’ health, safety, and privacy.

There are real harms in refusing to regulate health chatbots like ChatGPT Health. Chatbots used to diagnose or otherwise advise on a medical condition present a high risk of producing false negatives. Recent cases have demonstrated chatbots confidently (and incorrectly) citing blatantly falsified medical studies or generating misleading medical advice nearly 50 percent of the time. 

Unlike prior online medical resources like WebMD, health chatbots’ outputs are tailored to the user uploaded health records and thus framed as reasonably accurate applications of medical knowledge. But this accuracy is highly sensitive to users’ accurate self-evaluation of their symptoms and how they articulate them. Moreover, unfamiliar or socioeconomically disadvantaged users are more likely to prompt these errors, and have a strong financial incentive to take a false negative diagnosis at its word. In the context of healthcare, such reliance on incorrect or misleading outputs translates into delayed treatment and increased medical costs.

Despite these risks, the legislative activity surrounding chatbots remains inconclusive. While Congress has yet to pass comprehensive AI legislation, state legislatures are a hotbed of activity. More than 200 statutes currently exist across 48 states, many of which come into force this year. So far, these laws have been narrow in scope, mostly directed at transparency, consumer protection, and child safety. However, the future of many of these laws is uncertain. On Dec. 11, 2025, President Trump signed an executive order authorizing the Department of Justice to challenge state AI laws deemed problematic or onerous, a political move to reduce perceived regulatory burdens on AI innovation.

In this absence of legislative protections surrounding health chatbots, the FDA serves a crucial role based on its jurisdiction over medical devices.

Defining a Medical Device

When determining whether a product might be regulated as a medical device, a universal truth exists: If a product meets the definition of a medical device, it is subject to FDA jurisdiction. Further, the onus is on manufacturers to ensure compliance. 

The statutory definition of a medical device, per Section 201(h) of the Federal Food, Drug, and Cosmetic Act (FDCA) and as relevant to medical chatbots, is “an instrument, […] machine, […] or other similar or related article, including any component, part, or accessory, which is […] intended for use in the diagnosis[, …] cure, mitigation, treatment, or prevention of disease[.]” For products with multiple software functions, the FDA is empowered to “assess the impact” of the latter function on the overall device.  

When determining whether some product will be regulated as a medical device, the “intended use” is thus operative. If at least one function of ChatGPT Health is determined to have an intended use for the “diagnosis … or prevention of disease or other condition” — even if a separate function is intended to support a healthy lifestyle — then the FDA may determine the product under their regulatory authority. Under that perspective, ChatGPT Health should be implicated.

Determining Intended Use

OpenAI has been careful to disclaim certain uses of ChatGPT Health in its marketing to avoid implicating the medical device definition. The website announcing the platform states the following: “[ChatGPT] Health is designed to support, not replace, medical care. It is not intended for diagnosis or treatment.” 

Yet the FDA broadly construes intended use; the evidence used in ascertaining the objective intent of the manufacturer is not limited to promotional materials. Instead, the FDA may draw from any “other relevant source” — covering a broad array of activities and speech. Such available sources include knowledge of actual use by customers, the medical product’s design or composition, and the circumstances surrounding the distribution of the product. 

These “other relevant sources” are where ChatGPT Health is likely implicated as a medical device. Take for granted that consumers will use ChatGPT Health to try to avoid the time and expense of seeing a doctor for their lingering cough or the concerning blemish on their side. If ChatGPT Health lacks guardrails to prevent users misusing the product by seeking medical advice, that is an intentional feature of the product’s design and composition. Without such restrictive design — a standard we should hold health-related chatbots to given the high risk of incorrect or misleading medical advice — the FDA should be obligated to take action against ChatGPT Health as an unregulated medical device. For indeed, the intended use of ChatGPT Health encompasses symptom-based diagnosis and referral without real safeguards against the high likelihood of large-scale alternative use. 

ChatGPT Health should not be allowed to exist without guardrails or oversight. The FDA has the opportunity to intervene and protect the public from an unregulated medical device. Given the risk of real harm to users, the FDA should be obligated to do so.