OpenAI ChatGPT Health integrates Epic EHR patient data

Connecting ChatGPT Health directly to Epic’s EHR is a milestone in moving large language models from sidecar tools into the core of clinical workflows. Read-only access still avoids the scariest failure modes, but it gives the model enough context to summarize charts, surface trends, and draft documentation at the point of care. That is where the economic value in healthcare sits, and it is also where models get exposed to rich, longitudinal data.

From an AGI race perspective, this does two things. First, it accelerates the feedback loop between real-world clinical usage and model refinement. Even if OpenAI is not training on individual chats, usage patterns and error reports from thousands of clinicians will shape future model and product design. Second, it raises the bar for competitors: any lab that wants a share of healthcare now has to match not just model quality but also integrations, BAAs, and domain-specific safety evaluations.

The risk is that healthcare becomes an early battleground for overclaiming what these systems can safely do. OpenAI continues to insist ChatGPT is not for diagnosis, but as models grow more capable and tightly woven into records, the line between “assistive” and “decisive” behavior will blur. That tension will draw regulators, hospital risk teams, and malpractice insurers deeper into AI policy much sooner than they expected.