OpenAI Claims ChatGPT Health Can Outreason Clinicians

On Thursday, OpenAI is rolling out ChatGPT Health to logged-in U.S. users 18 and older, expanding a product that lets people connect medical records and health-tracking data just as the company makes an unusually aggressive claim about its models’ medical reasoning.

The feature will let users connect health provider records and services such as Apple Health, Weight Watchers, MyFitnessPal, and Function, then ask questions about that information inside ChatGPT, according to The Verge. The rollout covers web and iOS across free, Go, Plus, and Pro plans.

OpenAI’s models “are now capable of reasoning at levels that are better than clinician level,” Ashley Alexander, OpenAI’s vice president of health product, said during a briefing.

That line is the center of the story. ChatGPT Health is not just another health feature. It gives OpenAI a bigger role in how people interpret lab results, post-visit notes, medications, and fitness data, while the company is still drawing boundaries around what its chatbot should and shouldn’t say about care.

Thursday rollout expands ChatGPT Health to more U.S. users

ChatGPT Health has been framed by OpenAI as a dedicated experience for health information. With this release, OpenAI is letting more eligible U.S. users ask questions about connected medical information in ChatGPT.

That broader access matters. More availability usually means more use. It also means health context can become part of more user interactions when people choose to connect data, though OpenAI says ChatGPT will ask permission before using health information to personalize a response.

OpenAI says conversations are encrypted, and The Verge reports that ChatGPT will tell users “all conversations are encrypted at rest and at transit,” while information connected in Health “receives additional encryption protection.”

Users can review and change which health data they connect. The Verge lists several categories: post-visit notes, notes from a care team, lab results, and medications.

OpenAI’s own January 7 announcement framed Health as a dedicated experience for combining health information with ChatGPT. The company said then that health is one of ChatGPT’s most common use cases, with “over 230 million people globally” asking health and wellness questions each week, according to OpenAI.

That number explains the urgency. OpenAI isn’t trying to create demand from zero. It is formalizing behavior that already exists.

Clinician-level claim sharpens the stakes around ChatGPT Health

OpenAI’s public positioning includes a caveat. The company says ChatGPT Health “supports, not replaces, professional care.” Its January product post also says Health “is not intended for diagnosis or treatment.”

Those disclaimers now sit beside a much stronger marketing-adjacent claim: that OpenAI models can reason above clinician level. When pressed, Karan Singhal, OpenAI’s health lead, appeared to soften that framing.

Singhal said he would “temper” the claim that the models are reasoning at better levels, while saying “there have been individual studies that have been pointing in that direction,” including one from Harvard and Stanford.

That distinction is not semantic. “Individual studies” can show strong model performance on a defined benchmark or task. They don’t automatically prove reliable performance across messy patient situations, incomplete records, ambiguous symptoms, medication interactions, or emergencies.

XOOMAR analysis: The tension is clear. OpenAI wants users to trust ChatGPT with personal health context, but its own health lead is careful not to overstate what the evidence proves. The product claim and the safety disclaimer are pulling in different directions.

For readers tracking AI deployment risk beyond health, related XOOMAR coverage on OpenAI models breaching Hugging Face during a cyber test and AI hiring bias turning random noise into hidden job rules points to the same pressure point: model behavior matters most when outputs touch real-world decisions.

The supplied materials do not cite any regulatory action tied to ChatGPT Health. They do, however, show why scrutiny will focus on a narrower set of questions: what the system says when users ask about symptoms, treatment tradeoffs, medications, urgent warning signs, and whether to seek care.

January waitlist becomes a full U.S. release with more data controls

OpenAI introduced ChatGPT Health on January 7 as a dedicated experience and opened a waitlist. The broader U.S. rollout now expands that earlier test to eligible logged-in users across the company’s main consumer plans.

The release centers on allowing connected records and wellness data to inform ChatGPT responses. OpenAI says users can connect medical records and wellness apps so ChatGPT can ground responses in personal data.

Its Help Center says eligible users can connect Apple Health, supported U.S. provider portals, One Medical, and Function Health, with availability varying by provider and app.

The key operational question is data control. OpenAI says Health has added privacy protections, including encryption and isolation. Its January post says conversations in Health are not used to train its foundation models.

That claim will matter to users deciding whether to connect sensitive records. It also creates a practical test for OpenAI: privacy controls need to be understandable at the moment users are granting access, not buried after the fact.

Useful checks for users before connecting data:

• Scope: Review which records, apps, and categories you’re sharing.
• Permission: Watch when ChatGPT asks to use Health data in a response.
• Deletion: Confirm how to remove connected accounts or delete health chats.
• Care boundary: Treat the product as OpenAI describes it, support for care, not a replacement for clinicians.

The next proof point is evidence that matches the size of the claim

OpenAI has given ChatGPT Health a broad consumer launch and a bold performance claim. Now it has to prove both are safe enough for routine use.

The immediate evidence gap is not whether OpenAI models can perform well in selected studies. Singhal says some studies point in that direction. The harder question is whether those results hold when users bring fragmented records, vague symptoms, and high-stakes questions into an open-ended chatbot.

OpenAI’s next decision point is how much more evidence it publishes around ChatGPT Health’s performance and failure modes. The company has already said the product supports professional care and is not intended for diagnosis or treatment. The watch item is whether the user experience keeps that boundary clear when people ask medical questions.

If ChatGPT Health stays within those limits, it could make scattered health information easier to understand. If the product overreaches, the consequences won’t feel abstract. They’ll be personal, immediate, and difficult to unwind.

Impact Analysis

• ChatGPT Health could make AI a more central tool for interpreting medical records, lab results, and fitness data.
• OpenAI’s claim that its models can reason above clinician level raises major trust, safety, and accountability questions.
• Broader access for U.S. users increases the importance of privacy protections around sensitive health information.