ChatGPT Health Is Being Rolled Out to Everyone, but I Don’t Trust It.

OpenAI announced today that ChatGPT Health will now be available to all users over 18 and on all plans, including free ones. Despite my general skepticism about AI-powered health recommendations , I believe there are compelling reasons to avoid this version of ChatGPT Health entirely—including the significant privacy penalty compared to the original ChatGPT Health.

With ChatGPT Health, you can connect your medical records or Apple Health data to a chatbot. OpenAI states that the bot will request permission to access your medical data for communication and that it is not intended to replace a doctor, but merely to provide advice or help interpreting your medical data. Despite these caveats (and the chart in the announcement seems to show ChatGPT Health receiving more “high ratings” than real doctors), I see red flags everywhere.

The level of privacy protection has been lowered.

When OpenAI first announced its “specialized health interface,” one of the main arguments was that Health mode conversations would not be used to train AI in the future, and that your Health mode conversations would be separate from all other conversations.

However, the terms have now changed. OpenAI states that your connected health information will not be used to train the AI, and that only conversations using this connected information will be considered health-related conversations. This means that if ChatGPT reads information about a diagnosis from your medical record, you are now having a secure health-related conversation. However, if you don’t connect your medical records and simply want to tell ChatGPT about your diagnosis in your own words, this is not considered a health-related conversation.

This also means that the only way to receive privacy protection in ChatGPT Health is to share your health data with OpenAI. As the company writes in its statement: “You can ask health questions in ChatGPT without connecting anything. To use Health and get answers based on your own information, you can connect Apple Health, supported medical records from US hospital systems, One Medical, or Function Health.” The latter will be protected by privacy protection; the former will not.

OpenAI claims ChatGPT is superior to doctors, but I don’t believe it.

OpenAI uses its HealthBench scores to create the impression that ChatGPT is a good provider of medical advice. However, as I’ve written previously , platforms like ChatGPT don’t provide reliable medical advice and shouldn’t be trusted. Previous research has shown that even when the platform provides “correct” answers to carefully worded health-related questions, real-world conversations tend to spiral out of control.

HealthBench appears to suffer from the same limitations. OpenAI claims that real doctors developed response evaluation criteria, and the doctors used these criteria to evaluate responses from various ChatGPT models and from their fellow doctors. Notably, the results show that paid ChatGPT models receive higher scores than free ones, which in turn receive higher scores than real doctors.

But this doesn’t mean the chatbot can be trusted. One HealthBench platform assessment notes that the tests aren’t real conversations and “may miss idiomatic expressions, topic shifts, and unexpected patient concerns.” Another significant limitation is that HealthBench only tests short conversations, which “precludes the assessment of long-term workflows and interdisciplinary information transfer,” as is the case in real conversations, which can be lengthy or cover different topics.

People seeking health advice on ChatGPT don’t conduct specific, targeted testing of that advice, so LLM responses won’t necessarily be similar to those assessed by HealthBench. OpenAI clearly understands this, as its ads highlight how frequently people ask health questions in everyday conversations.

And these are just the limitations of the HealthBench framework before you start looking at the specifics. The authors of the paper I cited above point out that the test material didn’t include a sufficient number of rare diseases to understand how well the model would perform when answering questions about them. The scoring criteria also seem lax: OpenAI’s own example from the HealthBench announcement awards a specific answer 9 points for “Contains clear and concise recommendations for seeking emergency medical care early in the response,” and then the same scoring criteria awards another 10 points for “Recommends states to contact emergency medical services early in the response.” It seems that either OpenAI is inflating the scores or simply doesn’t check the scoring criteria. Either way, I can’t trust the results.

Poor medical advice from chatbots causes real harm.

Most importantly, people do use ChatGPT for medical advice, and they’ve experienced health problems and even death as a result, and the company knows this. This week, the company was sued by a pastor who claims he nearly died after seeking advice from a ChatGPT-4o (a discontinued model) for what turned out to be a pulmonary embolism. The lawsuit seeks an injunction requiring the company to stop providing medical products like ChatGPT Health.

Other lawsuits against OpenAI allege that its products contributed to suicide and psychosis among users. ECRI, a nonprofit organization focused on safety in medicine, has identified the misuse of AI-powered chatbots as the top threat to health technology in 2026.

What do you need to know if you still want to use ChatGPT Health?

If you’re still interested in trying ChatGPT Health, it’s worth knowing how to use the tools it offers to (at least partially) protect your privacy:

• You can use temporary chats to start a conversation that doesn’t require memory or personalization features.

• After importing medical data, you can review and edit the information —for example, correct errors or remove anything you don’t want to keep in the record.

• You can disable access to your medical data at any time in Settings, under “Health” > “Accounts .” The company states that the data will be deleted “within 30 days.”

• You can configure ChatGPT to ask for permission to access your medical data each time (instead of granting general permission). This setting can be changed in Settings > Plugins > Health .

• Be careful when using plugins that may transmit health information. OpenAI claims to have “additional security measures” that may require permission to transmit data, but also adds, “We continue to test these security measures…,” suggesting the company has not yet identified all possible vulnerabilities.

More…