I. Glenn Cohen

In January 2026, OpenAI launched ChatGPT Health, a product that allows patients to sync their electronic health records and wellness app data directly with a large language model and receive personalized responses to health queries. Microsoft Copilot has introduced a similar functionality for its LLM and others are entering the market Patients have had access to an AI chatbot offering medical information since ChatGPT's initial release but can now upload the entirety of their medical record to a commercial Large Language Models (LLM). This includes clinical data that clinicians would never share unfiltered with a consulting colleague. The result is not the democratization of medical knowledge, but the democratization of medical data, including the stigmatizing language, diagnostic errors, and institutional biases embedded in the record itself. In this Viewpoint, we evaluate some of the potential benefits and also discuss the attendant risks of privacy violations, discrimination, and the exacerbation of health disparities that may accompany the unfiltered upload of EHRs.

The Genetic Information Nondiscrimination Act (GINA) became law almost two decades ago, when genomic medicine was still in its infancy. One reason for its passage was to ensure that individuals and society would reap the benefits of emerging advances in genetic medicine, and would be able to benefit from genetic testing and research without fear of employment or health insurance discrimination. Since then, genomics has matured into a complex probabilistic science that increasingly allows for individualized estimates of genetic risk derived from large-scale population studies. Polygenic risk scores (PGSs), which provide genome-wide estimates of disease liability and may help indicate effective preventive care for an individual, raise new benefits but also concerns. PGS testing may become common in clinical practice, particularly to mitigate common complex diseases such as cardiac conditions and cancer. But are existing antidiscrimination protections adequate for a world where polygenic risk scoring is the norm? In this paper, we consider how existing laws apply and whether new legal and policy approaches are needed to support widespread, beneficial clinical use of PGSs. We also propose avenues for potential action by policymakers.