Microsoft launches Copilot Health to synthesize medical records and fitness data with AI

Microsoft Launches Copilot Health to Synthesize Medical Records and Fitness Data

Microsoft has unveiled Copilot Health, an AI tool designed to aggregate and summarize medical records, health history, and fitness data from multiple sources. The company claims the tool can "turn them into a coherent story" to help users understand their health holistically and prepare better questions for healthcare providers.

Data Integration and Sources

Copilot Health integrates data from more than 50 fitness and wearable devices including Apple Watch, Oura, and Fitbit. Through HealthEx, it accesses medical records—including visit summaries, medication details, and test results—from more than 50,000 hospitals and healthcare provider organizations in the US. The tool can also pull lab test data from Function with user authorization.

Microsoft notes that across its consumer AI products (Copilot and Bing), users ask more than 50 million health-related questions daily, indicating the scale of demand for such functionality.

Clinical Validation and Safety Claims

Microsoft claims Copilot Health was developed with input from its clinical team and more than 230 physicians from dozens of countries. The tool has achieved ISO/IEC 42001 certification, which the company describes as "the world's first standard for AI management systems."

The company emphasizes that Copilot Health explicitly does not diagnose, treat, or prevent diseases and is not a substitute for professional medical advice. Microsoft says it has incorporated "credible health organizations across 50 countries" into its training data, verified by its clinical team using principles from the National Academy of Medicine. Responses include citations and answer cards from Harvard Health.

Privacy and Data Handling

Microsoft states that Copilot Health operates in a "separate, secure space" within the Copilot app with dedicated access controls and safety measures. The company claims data is encrypted both at rest and in transit, and that it will not use Copilot Health information to train its models. Users can delete their information and revoke app access to health records and wearables at any time.

Availability and Pricing

Microsoft has opened a waitlist for Copilot Health, initially available in English in the US for users 18 and over. The tool will be free during an initial trial period. According to The New York Times, Microsoft plans to charge for access via subscription, but the company has not disclosed pricing details. The company is working on additional language, voice options, and international availability.

Broader Healthcare AI Context

Copilot Health enters a growing market of AI health tools. Amazon recently expanded its Health AI tool beyond One Medical, now available on Amazon's website and app, with Prime members able to chat with One Medical providers. OpenAI is testing ChatGPT Health, and Anthropic offers healthcare-focused tools.

Critical Considerations

While AI aggregation of fragmented health data could benefit users facing barriers to affordable healthcare, significant risks exist. AI hallucinations and incorrect medical advice remain documented concerns. Studies have shown LLM-based tools can provide incorrect medical information, and there is risk that such systems could downplay or exaggerate health risks. Entrusting comprehensive medical data to a chatbot differs fundamentally from using wearables for basic monitoring.

What This Means

Microsoft is positioning itself in the healthcare AI market with a tool focused on data synthesis rather than diagnosis. The free trial period represents a common strategy for health-related AI adoption, with monetization deferred. However, the lack of disclosed pricing and the gap between clinical validation efforts and the inherent unpredictability of large language models remain significant uncertainties. Real-world performance will depend heavily on how well the tool's summarization capabilities translate to clinical utility without introducing dangerous errors.