ChatGPT health records feature connects Apple Health data in US rollout
OpenAI is letting US users connect Apple Health and medical records to ChatGPT as a new lawsuit tests its medical-safety claims.
By Dominic Okoye · Staff Writer
· 3 min read
OpenAI is rolling out a ChatGPT health records feature that lets eligible users connect Apple Health data and supported medical records to the chatbot. The company did not announce a separate price for the feature, which is available to logged-in US users aged 18 and older on iOS and the web across Free, Go, Plus and Pro plans.
The move extends OpenAI’s push into healthcare data, a field where accuracy, liability and privacy constraints are higher than in most consumer AI use cases. OpenAI says the feature, called Health in ChatGPT, can use information a person chooses to share from Apple’s Health app, including medications, lab results, recent visits, sleep, activity and other stored health data, to provide more personalized responses in health-related chats.
What is Health in ChatGPT?
Health in ChatGPT is an integration with Apple’s Health app that gives ChatGPT access to selected health information after a user grants permission. Apple Health collects data from an iPhone, Apple Watch and compatible fitness or health apps, and it can also store user-entered information such as medications, allergies and medical conditions.
OpenAI said it built the broader integration after testing a dedicated health experience inside ChatGPT earlier this year. According to the company, users often skipped that separate area, with more than 70% of health-related conversations taking place outside the dedicated health space.
OpenAI says it will not use connected health data to train its foundation models or to serve ads. The company also says ChatGPT can still make mistakes and should not replace medical professionals, even as it claims to have worked with hundreds of physicians to test the feature for performance and safety.
Why is the timing drawing scrutiny?
The launch comes one day after a Florida man sued OpenAI in San Francisco Superior Court, alleging ChatGPT gave him dangerous medical recommendations as his pulmonary embolism worsened. The plaintiff claims the chatbot repeatedly told him his symptoms were not serious and encouraged him to stay home, after which he was hospitalized and now faces years of physical and psychological recovery.
An OpenAI spokesperson told The New York Times that ChatGPT’s terms of service state the product is not intended for medical diagnosis or treatment. The spokesperson also told the newspaper that OpenAI’s models have improved since the incident at issue in the lawsuit. OpenAI made a similar point to The Register.
In an email to The Register, an OpenAI spokesperson said, “ChatGPT is not a doctor and should never be used as a substitute for medical care, diagnosis, or treatment.” The spokesperson also said treating chatbots as the full explanation for people’s medical decisions or outcomes “oversimplifies a much bigger challenge” and could interfere with access to tools that may aid users in health-related tasks.
More context from personal records could help in some conversations, especially where prior diagnoses, medications or test results are relevant. It does not solve the core risk that a general-purpose model can give incorrect advice with confidence.
Recent research and oversight findings have kept that risk in view. One study found AI chatbots were no better than conventional online resources, including search engines, at helping people make health decisions. Another found large language models failed at early differential diagnosis in more than 80% of cases. In Canada, Ontario’s Auditor General found AI note-taking tools used by doctors introduced medication errors, left out critical details, added incorrect information and hallucinated details not discussed in appointments.
OpenAI says it uses dedicated training for health conversations and that its models have made gains in recognizing when urgent care may be needed, asking for relevant context, explaining uncertainty and making complex information easier to understand. Users, clinicians and regulators will now test those claims against a product that asks for more intimate data than a typical chatbot prompt.
This story draws on original reporting from The Register.
