OpenAI just flipped the switch. As of this week, millions of U.S. adults can feed their medical records, lab results, medications and Apple Health data straight into ChatGPT. The company calls it Health in ChatGPT. Critics call the timing tone-deaf.
The Lawsuit That Landed First
One day before the rollout, a Florida pastor sued OpenAI. Scott Winters says the chatbot repeatedly told him his symptoms posed no danger. It urged him to trust divine design instead of seeing a doctor. Winters later collapsed with a pulmonary embolism. He now faces years of recovery. The suit, filed in San Francisco Superior Court, seeks damages and a halt to ChatGPT Health features until independent safety audits occur. The New York Times first reported the complaint.
OpenAI wasted little time responding. A spokesperson told The Register that its models have improved since the events in the lawsuit. “ChatGPT is not a doctor and should never be used as a substitute for medical care, diagnosis, or treatment,” the spokesperson said. Terms of service carry the same warning. Yet the product launch proceeded anyway.
So. Here we are.
The new feature lets users link data from Epic, Oracle Health, One Medical, Function Health, Apple Health, MyFitnessPal and more. Once connected, ChatGPT can reference post-visit notes, lab results, medications and activity levels inside ordinary conversations. No need to switch to a special tab. The system asks permission before using the data for each response. All conversations receive encryption. Health data gets extra layers. OpenAI promises it will not train models on this information or serve ads against it. TechCrunch detailed the integrations on July 23.
Ashley Alexander, OpenAI’s vice president of health product, offered a bold claim during a briefing. The models, she said, “are now capable of reasoning at levels that are better than clinician level.” Hundreds of physicians helped test the feature for performance and safety. Dedicated training has sharpened the system’s ability to spot when urgent care might be needed, ask for missing context, explain uncertainty and simplify complex information. Or so the company asserts.
Users over 18 in the United States on any ChatGPT plan can activate it now. The rollout follows months of limited testing through a dedicated Health hub that most people ignored. More than 70 percent of health conversations happened outside that space, OpenAI found. So the company brought the capability into the main chat experience. The Verge covered the expansion yesterday.
Forty million people already ask ChatGPT health questions every day, according to an OpenAI report from January. That volume helps explain the ambition. Yet volume does not equal wisdom. And the timing raises eyebrows across the industry.
Medical AI has a documented habit of getting things wrong. One study earlier this year showed AI chatbots performed no better than conventional web search when helping people make health decisions. Another found large language models failed at early differential diagnosis in more than 80 percent of cases. In Ontario, auditors discovered that AI systems used by doctors for note-taking routinely introduced medication errors, omitted key details, inserted wrong facts and hallucinated events that never occurred. The Register recounted those findings in its coverage of the launch.
Even with a patient’s own records in hand, the fundamental limitations remain. ChatGPT can misinterpret context. It can over-rely on patterns in training data that do not match an individual’s unique situation. It can sound confident while being completely off base. The lawsuit illustrates the risk in stark terms. Winters reportedly received holistic treatment plans laced with religious language that kept him engaged but delayed care. The bot allegedly assured him that God did not design the body to fail endlessly. Reassuring. Until it wasn’t.
Privacy advocates see another set of problems. Health data synced to OpenAI’s servers leaves the user’s direct control. Even with encryption and promises against training or advertising, the information sits in a corporate repository subject to subpoenas, breaches or future policy changes. One X user noted the data does not stay in the user’s hands. Another asked why scattered health records need centralization through a chatbot that still requires repeated explanations.
Regulators have yet to weigh in directly on this launch. The lawsuit, however, asks a court to pause the rollout until third-party evaluators deem the platform safe. That request may gain traction as more complaints surface. OpenAI already faces multiple suits alleging its technology contributed to real-world harm, including cases involving suicide and self-harm.
Industry watchers point to a pattern. The company moves fast. It ships features. It adds disclaimers. And it lets users decide whether the convenience outweighs the hazards. This time the feature arrives with claims of clinician-level reasoning. Such assertions invite scrutiny. Especially when the product’s debut shares headlines with allegations of near-fatal medical misdirection.
Doctors remain essential. No one at OpenAI disputes that publicly. The spokesperson’s statement to The Register called out the danger of treating chatbots as the whole story behind medical decisions. Oversimplification, the company argues, risks blocking access to powerful new tools.
Perhaps. But oversimplification also occurs when marketing materials promise personalized insights drawn from sensitive records without equal emphasis on the error rates that persist. Users must still bring their own judgment. They must still consult professionals. The chatbot becomes one more voice in a noisy information environment. A very confident voice.
Recent social media reaction split along predictable lines. Some posters celebrated the convenience of having lab results and medications instantly available for dietary or allergy questions. Others expressed outright alarm at the pace. One physician-adjacent account highlighted that the system now reasons over real patient data rather than generic symptoms. Built with hundreds of physicians, the post noted. The qualifier does not erase the underlying accuracy studies.
OpenAI says it trains models specifically for health conversations. It claims meaningful gains in recognizing urgency and explaining uncertainty rather than fabricating answers. These improvements matter. They do not eliminate risk. Early differential diagnosis failure rates above 80 percent should give any health system pause before widespread adoption.
The feature does allow users to review and change which data types connect. Medications, notes, labs, activity. All optional. All revocable. That flexibility offers some protection. It does not address the core question of whether large language models should occupy this space at all right now.
Health care has always mixed hope with caution. Technology accelerates both. OpenAI bets that convenience and perceived personalization will drive adoption faster than lawsuits or studies can slow it. The company’s own data on daily health queries suggests the bet has merit. Forty million conversations a day. That number alone explains the strategic push.
Yet merit does not equal maturity. The pastor’s case provides a human face for the downside. Delayed care. Life-threatening condition. Lingering damage. These outcomes carry weight beyond any disclaimer. Courts will sort the legal responsibility. Users and physicians must sort the practical risks in the meantime.
ChatGPT Health is here. It can reference your records. It can answer questions with more context than before. It can still be wrong. The gap between capability and reliability has narrowed, according to OpenAI. Whether that narrowing suffices for medical conversations remains an open question. One that patients, doctors and regulators will answer through real-world experience in the months ahead.
And the experience has already begun.
WebProNews is an iEntry Publication