OpenAI opens ChatGPT Health to the whole U.S., claiming it surpasses clinicians but also gets sued

OpenAI opens ChatGPT Health to users across the U.S., claiming its reasoning surpasses that of clinical physicians, but the responsible executive retracts on the spot; in the same week, a pastor sued it for giving dangerous advice that caused him to delay seeking medical care, directly pointing to unlicensed practice and placing user engagement above safety.
(Background: OpenAI launched a medical version of ChatGPT for Clinicians, with 99.6% accuracy assisting human doctors (free and open in the U.S.))
(Additional context: Don’t hand your medical records to a chat robot? The privacy gamble under ChatGPT Health’s medical ambitions)

This Thursday (23), OpenAI rolled out ChatGPT Health to users nationwide and laid down a firm statement: Ashley Alexander, Vice President of Health Products, said at a briefing that the company’s model “now has reasoning capabilities that have surpassed the level of clinical physicians.”

But in the same week, a pastor in Florida sued OpenAI in the Superior Court of San Francisco, alleging that ChatGPT provided “extremely dangerous medical advice,” causing him to delay getting treatment; a blood clot later worsened into a massive pulmonary embolism.

From being for doctors to opening up to everyone

This update is not a new feature—it expands the battlefield. On April 23, OpenAI launched ChatGPT for Clinicians, limited to verified medical professionals, and simultaneously released HealthBench Professional evaluations targeting that niche group of doctors. Three months later, the threshold was removed entirely: anyone aged 18 or older who logs into ChatGPT can use it, covering all tiers—free, Go, Plus, and Pro—first rolling out on the web and iOS.

Supporting this opening is a new model published a few weeks ago, GPT-5.6 Sol. OpenAI says it is the company’s “strongest model to date in the health domain,” and after ChatGPT Health, it will run directly on this model. On the feature side, integration was also done: previously, users had to jump to a separate ChatGPT Health tab to ask questions related to health data; now they can ask directly in the main chat window. In other words, health-data queries become part of everyday conversation, without needing to switch interfaces.

Before use, ChatGPT will first request authorization to customize answers using health data. OpenAI emphasizes that “all conversations are encrypted both during transmission and static storage,” and data linked into Health “receives additional encryption protection.” The encryption protocol sounds stringent, but the real risk has never been whether data leaks; it is whether the system, after turning the data into advice, leads users to trust it.

Users can review and adjust the types of data they link, including post-visit summaries, care team notes, lab reports, and medication history. Data sources include not only medical institutions, but also third-party services such as Apple Health, Weight Watchers, MyFitnessPal, and Function.

The pastor’s lawsuit punctures the same sentence

Right before the rollout, a complaint pulled the “surpassing doctors” claim back to reality. Plaintiff Scott Winters, 55, a former pastor in Florida, filed suit on July 22 in the Superior Court of San Francisco. He alleges that ChatGPT-4o repeatedly downplayed his symptoms, advised him to stay at home on the lounge chair, discouraged him from seeing a physician, and ultimately allowed the blood clot inside him to worsen into a massive pulmonary embolism.

The most striking part of the complaint accuses ChatGPT of using his own religious background to persuade him—packaging the chat robot’s advice in language of faith, making him more comfortable waiting at home rather than going to the emergency room. A patient who should have been treated as a warning sign and referred was calmed all the way to worsening by a trusted tool—this is the grayest and most dangerous zone between “assisted decision-making” and “replacement of judgment.”

The complaint raises 8 causes of action at once. Beyond negligence, violations of California consumer protection laws, and privacy laws, it also directly alleges that OpenAI constitutes unlicensed medical practice and puts user engagement above safety. So-called unlicensed medical practice, in plain terms, is that the plaintiff argues ChatGPT’s advice goes beyond the scope of “information reference” and effectively replaces doctors in making diagnosis and treatment decisions—without a license and without being held responsible for medical outcomes.

However, which specific model the pastor used is currently unclear.