As AI-assisted health navigation becomes normalized, the decision to hand an entire electronic health record to a large language model is no longer hypothetical — it is already happening. Understanding the tradeoffs between personalized insight and structural risk matters enormously for anyone managing a chronic condition, tracking biomarkers, or seeking a second opinion in an era of fragmented care.
This JAMA Viewpoint examines what occurs when patients upload unfiltered EHR data to consumer-facing large language models. The core tension is that comprehensive data input theoretically improves AI-generated guidance — the more context a model has, the more tailored its outputs. Yet that same comprehensiveness exposes sensitive diagnostic codes, mental health records, genetic findings, and medication histories to platforms with variable data retention, third-party sharing agreements, and opaque training pipelines. The authors flag discrimination risk as a specific concern: insurers, employers, or other actors could potentially access inferred health status if data governance fails.
From a broader research perspective, this Viewpoint arrives at a pivotal moment. Large language models have demonstrated genuine clinical reasoning capacity in several peer-reviewed benchmarks, yet regulatory frameworks governing patient data shared voluntarily with non-HIPAA-covered AI platforms remain underdeveloped. HIPAA protections do not automatically extend to consumer AI tools, meaning patients who upload their own records may inadvertently step outside legal safeguards designed to protect them. The health disparity angle is underexplored in public discourse: populations with the most to gain from AI-assisted navigation — those with limited access to specialist care — may also be least equipped to evaluate platform trustworthiness. This piece is incremental rather than paradigm-shifting, but its JAMA platform amplifies a warning that clinicians, patients, and policymakers should take seriously before AI health companionship scales further.
