Context-Aware Health Intelligence
OpenAI has integrated personal health data directly into the ChatGPT experience, moving away from a siloed "health-only" interface. By connecting Apple Health and supported medical records (including One Medical and Function Health), users can now ground their conversations in their own medical history, lab results, and wearable data. This allows the model to perform complex tasks like comparing new lab results against historical data, summarizing changes since the last appointment, or providing dietary advice based on specific health restrictions.
Privacy and Security Architecture
To address the sensitivity of medical information, OpenAI has implemented a "privacy-first" design:
- No Model Training: Data connected through the Health feature is explicitly excluded from training foundation models or targeting advertisements.
- Granular Control: Users must grant permission for ChatGPT to access their data. By default, the system prompts for approval before each access, though users can toggle this to "always allow."
- Data Lifecycle: If a user disconnects an account, the synced data is purged from OpenAI’s systems within 30 days.
- Encryption: All health data is encrypted both at rest and in transit, with additional layers of protection applied to connected medical records.
Model Performance and Clinical Evaluation
This feature leverages GPT-5.5 Instant and GPT-5.6 Sol, the latter being optimized for complex reasoning across health-related tasks. OpenAI developed "HealthBench Professional" to evaluate model performance, working with hundreds of physicians to define rubrics for accuracy, safety, and communication. While these models demonstrate improved capabilities in explaining medical notes and recognizing when professional care is required, OpenAI emphasizes that the tool is intended to support—not replace—the judgment of qualified medical professionals.