OpenAI unveils ChatGPT Health space for secure data

OpenAI has launched ChatGPT Health, a dedicated version of its chatbot that sits inside ChatGPT as a separate space for health and wellness queries and integrates with medical records and consumer health apps.

The new service targets people who already use ChatGPT for questions about symptoms, test results and lifestyle changes. OpenAI said health-related use is one of the chatbot's largest categories, with more than 230 million people globally asking health and wellness questions each week based on de-identified analysis of conversations.

ChatGPT Health operates as a distinct environment within ChatGPT with its own storage, memories and controls. The company said it keeps sensitive health data compartmentalised from other chats and applies additional privacy and security safeguards on top of those used across the wider product.

Users can connect electronic medical records and wellness applications so that responses reflect their own health data. Early integrations include Apple Health, b.well's health data network in the US, and consumer apps such as Function, MyFitnessPal and Weight Watchers, alongside services like AllTrails, Instacart and Peloton.

Health conversations sit inside the main ChatGPT history for ease of navigation. OpenAI said the underlying information remains contained within the Health space and does not flow into non-health chats. Health memories remain separate. Users can view or delete these within Health or through the Personalization settings.

The company said conversations and files across ChatGPT are encrypted at rest and in transit. ChatGPT Health adds purpose-built encryption and isolation for health data. OpenAI said it does not use Health conversations to train its foundation models.

Users can turn on multi-factor authentication as an additional access control for their accounts. They can also disconnect health apps and revoke access to medical records at any time through the Apps section of settings. Apps inside Health must meet OpenAI privacy and security requirements and undergo a specific security review for inclusion.

The product launches initially to a limited set of early users. People with ChatGPT Free, Go, Plus and Pro plans outside the European Economic Area, Switzerland and the United Kingdom can join a waitlist. OpenAI plans to expand access on the web and iOS in the coming weeks. Medical record integrations and some app connections are currently available only in the US, and syncing with Apple Health requires iOS.

Clinical framing

OpenAI positions ChatGPT Health as a tool for everyday questions and preparation around medical care. It states that the product is not intended for diagnosis or treatment and is designed to support rather than replace clinicians.

Users can ask the system to explain recent test results, prepare for doctor's appointments, outline diet and workout approaches, or set out trade-offs between insurance options based on their healthcare usage. The company said Health can also help users identify trends over time, such as changes in cholesterol or sleep, by referencing connected data sources.

ChatGPT may suggest that users move a health-related conversation into the Health space, which applies the added protections. When relevant, it can draw on context from non-health chats, such as a recent house move, but the flow of information does not run in the other direction. Non-health chats cannot access files, conversations or memories that sit in Health.

People can upload photos and files, use voice mode and dictation, and invoke specific apps for tasks such as turning meal plans into shoppable grocery lists or pairing activity data with exercise recommendations. Custom instructions in Health allow users to set preferences on topics, tone and focus, and these instructions apply only within the Health environment.

Physician input

OpenAI built ChatGPT Health in collaboration with a network of practising physicians from multiple countries and specialties over a two-year period. The company said more than 260 physicians, with experience in 60 countries, have provided feedback on model outputs over 600,000 times across 30 areas of focus.

The clinicians advised on when answers should encourage follow-up with a doctor, how information should be phrased for clarity without oversimplification, and how the system should handle situations that require higher safety thresholds. Their feedback informed the model that underpins ChatGPT Health and the way it responds to different types of queries.

OpenAI uses an evaluation framework called HealthBench to assess the system. The framework is based on physician-written rubrics and clinical standards. It focuses on factors such as safety, clarity, appropriate escalation and respect for individual context, rather than exam-style questions or generic accuracy checks.

The company said this assessment approach reflects the tasks users are likely to perform with the product. These tasks include explaining lab results in accessible language, preparing questions for appointments, interpreting data from wearables and summarising care instructions after visits.

"This physician-led approach is built directly into the model that powers Health, which is evaluated against clinical standards using HealthBench, an assessment framework we created with input from our network of practicing physicians," said OpenAI.

OpenAI plans to broaden the range of data sources that link into ChatGPT Health and expand the types of insight that the system can offer as it gathers feedback from early users.