It happens fast. One minute you’re asking about a weird rash. The next, an AI bot suggests a treatment that might be lethal. OpenAI has officially opened the gates.
ChatGPT Health is now live for every U.S. user over 18, across all subscription tiers. This isn’t just a beta test anymore. It’s a full-throttle rollout that lets anyone with an iPhone or a laptop plug their Apple Health data into a general chat window and ask, “What’s wrong with me?”
The timing is grotesque. This launch drops just a day after a Florida pastor sued OpenAI. The suit alleges the chatbot gave a near-fatal recommendation to skip seeing a doctor. The company ignored him then. Now it’s scaling the very feature that sparked the litigation.
The Data You’re Feeding the Machine
OpenAI claims it knows what you want. In early January, during limited testing, users were hitting the health hub 230 million times a week. That number jumped to 300 million before the doors were thrown open.
People are desperate for answers. They want to know if their food allergies will flare up. They want quick takes on weird symptoms. So, they connect everything.
The system now integrates data from Epic and Oracle Health hospital records. It pulls from One Medical. It syncs with Function Health and MyFitnessPal. All of it.
During the test phase, OpenAI watched 70% of queries happen outside the dedicated health hub. Users didn’t want a separate interface. They wanted context-aware answers in the main chat. They wanted the bot to know they’re diabetic while discussing a vacation itinerary.
How Well Does It Actually Work?
OpenAI is betting on GPT 5.6-Luna. Specifically, the smallest model in the latest release.
They say it outperforms GPT 5.15 on HealthBench. HealthBench is an open-source benchmark OpenAI created to grade LLMs on medical queries. If the numbers hold up, the smallest model is better at triage than the previous generation.
But there’s a catch.
OpenAI says it doesn’t use user data to train its models. It works with physicians to refine answers.
This sounds safe. It feels professional. It might even be true.
But here is the hard truth: AI is not a doctor. It’s a pattern-matching engine trained on the internet’s entire medical library, plus your personal Apple Health metrics.
The company maintains that these tools are “not intended for use in diagnosis or treatment.” They cite these disclaimers constantly. They told the New York Times they are making health-related answers “safer.” They urge you to verify information.
Verify. With a human.
Why This Won’t Stop
Other companies aren’t listening. Anthropic launched health features. Google launched health features. Both have been burned by similar errors. Both kept going.
Why? Because users keep clicking. Because 300 million health queries a week represent a massive shift in behavior. People trust the machine more than they trust the wait times at their primary care physician.
OpenAI knows this. That’s why they aren’t hiding behind a gated hub anymore. They want you to use their data in every query. They want the integration to be seamless.
Studies repeatedly show AI bots are unreliable for medical advice. Hallucinations are common. Confabulations happen. Yet, the rollout continues.
Health in ChatGPT is available to logged-in U.S. users on the web and iOS today. Free, Go, Plus, and Pro. All of them.
You have the tools. You have the data. You have the permission to ask.
But when the screen glows back at you with a suggestion to stop taking your heart medication because a subreddit post says otherwise… will you stop?
Or will you just close the tab and hope for the best?
