OpenAI makes ChatGPT Health available to all US users

OpenAI just pushed ChatGPT Health to every logged-in American over eighteen. The timing is either tone-deaf or calculated. Twenty-four hours earlier, a Florida pastor filed suit claiming the chatbot told him not to see a doctor — a suggestion that nearly killed him. The company's response: a terms-of-service disclaimer and a promise to make answers "safer." Then it shipped the product anyway.

Three hundred million health queries a week. That number climbed from 230 million in January. Users are not waiting for permission. Seventy percent of those queries already happened outside the dedicated health hub OpenAI built for them. People asked about symptoms, drug interactions, lab results — in the same window where they draft emails and debug code. The new feature simply admits reality: the hub was a fiction. Now the model can pull Apple Health exports, Epic records, Oracle Health data, MyFitnessPal logs, One Medical files, Function Health feeds — whatever the user connects — into any conversation. Ask about dinner ingredients and the bot cross-references your allergy list. That is the product.

OpenAI touts GPT 5.6-Luna. The smallest model in the latest release supposedly outperforms GPT 5.5 on HealthBench, an open-source benchmark OpenAI itself developed. Benchmarks the creator writes tend to reflect the creator's priorities. Independent replication matters more. The company also says it does not train on user data and consults physicians to improve outputs. Both claims are standard. Neither eliminates hallucination.

The terms of service have not changed. ChatGPT "is not intended for use in the diagnosis or treatment of any health condition." That sentence carries more legal weight than any benchmark score. It sat in the terms when the pastor asked his question. It sits there today. OpenAI cited it in court. It cites it now. The disclaimer is the product's true architecture.

Studies keep piling up. Large language models invent dosages, miss contraindications, confuse guidelines. They speak with the confidence of a board-certified Internist and the accuracy of a WebMD rabbit hole. That has not stopped Anthropic. It has not stopped Google. Med-PaLM, Med-Gemini, whatever the next codename — every lab wants the halo of healthcare revenue without the liability of a medical device license. The FDA watches. Class-action lawyers watch. Users proceed regardless.

The rollout covers Free, Go, Plus, and Pro tiers on web and iOS this week. Android follows. No age verification beyond a checkbox. No clinical oversight. No mandatory human-in-the-loop for high-risk topics. OpenAI says it wants people to verify information and seek professional advice. That is a hope, not a control. The interface encourages the opposite: instant, authoritative, personalized answers that feel like care.

Health data is the stickiest data. Once a user links Epic and Apple Health, switching costs explode. The moat deepens. OpenAI knows this. The 300 million weekly queries represent intent that no search engine, no patient portal, no nurse line captures. Monetization follows attention. Today the feature is free across tiers. Tomorrow it becomes the anchor for a Pro Health tier, an enterprise EHR integration, a pharma partnership. The disclaimer stays. The revenue grows.

Physicians will see more patients arriving with ChatGPT printouts. Some printouts will catch a missed interaction. Most will waste visit time debating a hallucinated side effect. The net effect on clinical workflow is negative until proved otherwise. OpenAI does not run that trial. It releases.

The pastor's lawsuit alleges negligence. Discovery will demand prompt logs, model versions, safety filtering rules, red-team reports. OpenAI will argue Section 230, argue the disclaimer, argue the user ignored the warning. The court will decide whether a chatbot that integrates your EHR and answers "should I go to the ER" with "probably not" functions as a medical device. That decision sets the boundary for the entire category.

Until then, 300 million queries a week keep flowing. The hub is dead. The disclaimer stands. The benchmark glows. The pastor recovers. The rollout continues.