OpenAI Connects ChatGPT to Epic Medical Records System for Millions of Patients

OpenAI has connected ChatGPT to Epic, the world's largest electronic health records system, giving authorized physicians access to a patient's medical history directly from within the chatbot. It's the company's latest step in its expansion into healthcare, following the earlier launch of the consumer-facing ChatGPT Health and the clinical ChatGPT for Clinicians.

The new feature lets physicians ask ChatGPT questions about a specific patient and receive summaries drawn from visit notes, lab results, specialist records and medication information. The system flags what has changed since the previous visit, surfaces the latest test results, tracks medication changes and highlights issues that need follow-up.

How the Epic integration works

The Epic connection operates in read-only mode, meaning ChatGPT can pull and analyze data from a patient's chart but cannot write new information into it. The approach is meant to limit the risk of erroneous data entry into a system used by hospitals and clinics worldwide.

ChatGPT can identify what changed since a previous visit, surface recent lab results, track medication changes and highlight unresolved follow-ups - OpenAI, description of the ChatGPT for Clinicians feature

Only authorized clinicians at facilities that have opted into the Epic integration can access the data. The company stresses that the tool is meant to support, not replace, a physician's work, and is not intended for diagnosis or treatment.

Testing with hundreds of physicians

Before rolling out the feature, OpenAI ran extensive testing with a global network of physician advisors from 60 countries and 26 medical specialties. So far, they have issued more than 700,000 ratings of the model's answers to questions reflecting real clinical situations.

In tests covering 27 clinical scenarios, such as visit preparation, medication review, treatment timelines and handoff summaries between physicians, 99.1 percent of the 4,363 assessed responses were rated safe. In a separate round of testing based on large US medical datasets, more than 93 percent of responses for each of the five connected data sources were rated as having good or better accuracy.

Three versions for different audiences

OpenAI is developing three health products in parallel. ChatGPT Health, launched in January 2026, is a section for everyday users, currently available only in the US for adults, and connects to Apple Health, MyFitnessPal and Peloton through a partnership with b.well, which operates one of the largest connected health data networks in the US. ChatGPT for Clinicians, which launched in April 2026, is aimed at physicians and is the one that has now gained the Epic connection. The third version, ChatGPT for Healthcare, serves researchers and administrators at medical facilities.

The company says health data is stored separately from other conversations and is not used to train GPT models. The system is also meant to offer dedicated encryption and memory isolation for sensitive medical information.

Data protection concerns

Not every expert views OpenAI's push into medicine uncritically. Andrew Crawford of the Center for Democracy and Technology points out that tech companies not covered by HIPAA (the US law that protects patient data at traditional healthcare providers) can handle health data under their own rules.

it puts sensitive information at real risk - Andrew Crawford, Center for Democracy and Technology

OpenAI responds to these criticisms with assurances of compliance with security requirements and stresses that the tool does not replace seeing a doctor. In its communications, the company repeatedly states that ChatGPT Health and related products are not meant for diagnosis or treatment, only for supporting patients and medical staff in everyday tasks.

For Polish healthcare facilities and doctors, the Epic integration has limited relevance for now, since the system dominates primarily in the US market. It does, however, point to where the market is heading: major AI providers want to plug directly into hospital and clinic infrastructure rather than just offer a general chatbot for symptom checks.

OpenAI has more than 230 million weekly users, the audience targeted by the consumer version of ChatGPT Health, making the company one of the biggest players trying to reach both patients and medical staff at once. The scale of testing, involving hundreds of physicians from dozens of countries, is meant to argue that the rollout is safer than a general chatbot without medical oversight.