ChatGPT Is Inside Epic Now. Nurses Say No One Asked Them If It’s Safe

Part of Nurse.org’s Nursing AI Watch, our ongoing investigation into how artificial intelligence is reshaping nursing practice. In this piece we look at ChatGPT’s arrival inside Epic, and the gap between the nurses who will use the tool and the physicians who were named in testing it.

ChatGPT has arrived inside the patient chart. UCSF Health is piloting OpenAI’s new Epic integration, which pulls the notes, the labs, the med list, and hands clinicians a summary of what changed and what matters.

Whether those summaries are safe was decided by 4,363 ratings from hundreds of physicians. In OpenAI’s own announcements, not one nurse is named among the people who tested it. The announcement also names health systems, physicians, and public data partners; nurses appear nowhere in it.

Here’s what OpenAI announced about the ChatGPT-Epic integration, what nurse leaders are saying, and what our research adds: at the pilot hospital, the nurses’ union contract gives them a right to a seat at exactly this kind of table.

Want to see more Nurse.org articles in your Google results? Add us as a preferred source.

On September 1, OpenAI announced that ChatGPT for Healthcare, its version for hospitals and health systems, now connects to Epic. Clinicians can pull an authorized patient’s chart into ChatGPT for a summary, or use ChatGPT without leaving the chart.

UCSF Health is the pilot, and there are seven launch partners in all: AdventHealth, Baylor Scott & White, Boston Children’s, Cedars-Sinai, HCA Healthcare, Memorial Sloan Kettering, and UCSF.

Three details matter most:

• It’s read-only. OpenAI’s release notes confirm it can summarize the chart but can’t write anything back. No charting, no orders, no help with the documentation load nurses actually want off their plates.
• You can’t just turn it on. It requires an enterprise workspace and a signed Business Associate Agreement, and it’s not available for individual accounts.
• It is not in front of 325 million patients. That viral number is Epic’s total footprint. This feature is live at pilot scale.

According to the OpenAI announcement, the plug-in is supposed to provide relevant clinical context for providers to have a quick summary snapshot of the patient’s case, including what’s changed since their last visit and any relevant medical updates from external sources, like a new relevant study or medication change, for instance.

In OpenAI’s own words, its physician network exists to “define, measure, and improve health responses” in ChatGPT, and the company credits that physician feedback with improving model behavior.

Who Tested ChatGPT in Epic, and Who Didn’t

OpenAI says hundreds of physicians across 60 countries have reviewed more than 700,000 responses. For the Epic integration, physicians scored 27 clinical use cases, and 99.1% of 4,363 rated responses were rated safe.

Those numbers deserve a closer look. “Rated safe” is a rating from a study the vendor designed and physicians scored. It also means about 1 in 100 responses wasn’t rated safe, and errors that slip through a chart summary tend to surface at the bedside, on a nurse’s shift. One of the 27 use cases is handoff summaries. Handoffs are typically nursing work.

And the pattern holds across the whole year. ChatGPT Health launched in January, built with more than 260 physicians. ChatGPT for Clinicians launched in April, reviewed by “hundreds of physician advisors.” Then the Epic integration in September. No nurses are named as evaluators or advisors in any of them. Nurses appear exactly once: nurse practitioners made April’s list of who’s allowed to use it.

Could nurses have been involved without being named? It’s possible, and worth saying plainly. But OpenAI isn’t vague about who did this work. In every announcement, the company describes its advisory network as physicians and says physicians did the evaluating. So either nurses weren’t part of testing a tool nurses will use, or they participated and went unmentioned in announcements that specify physician counts every time. Either way, the public record gives nurses no evidence they were represented.

Epic takes a different approach with its own AI. Epic’s assistant, Art, also summarizes charts, and its product page specifically says it can draft an end-of-shift note for nurses. We covered Epic’s AI charting tools and what they mean for nursing documentation. As of publication, Epic has not publicly commented on OpenAI’s announcement. The day after, its newsroom published stories about its own AI tools.

>>Listen to The Latest Nurse News Podcast

Nurse leaders spotted the gap within 48 hours.

Kathryn Shaffer, EdD, RN, MSN, CNE, Director of Innovation at Jefferson College of Nursing, wrote on LinkedIn that January’s launch involved 260 physicians and zero nurses, and that eight months later the tool is in the chart while the validation still doesn’t include the profession that lives in it. The graphic on her post said it in seven words: “Absent from the study, present for the consequence.”

Rebecca Love, RN, MSN, Nurse.org contributor and Love n’ Leary podcast host, opened her public post with “Yikes!” and wrote that the hope now rests with the humans on the front lines of care.

Bradley Goettl, DNP, Chief Nursing Officer of the American Nurses Enterprise, asked the operative question in a LinkedIn discussion of the news: was this full integration into the EHR operation, or a connection to personal health data specific to ChatGPT Health? It’s the right question, and the announcement doesn’t make the answer easy to find.

Nurses at at least two of the seven launch systems wrote AI rules into their union contracts before this tool ever arrived.

At UCSF, about 6,400 nurses are represented by the California Nurses Association under a systemwide UC contract running through January 2030. Article 42, titled New Technology, says clinical technology, including AI, must complement nursing judgment, not diminish it, and can’t replace the nurse’s role in care. It also gives nurses a seat: their Professional Practice Committee is entitled to participate in the “selection, design, building and validation” of new technology affecting nursing care, and the chief nursing officer must respond in writing to the committee’s recommendations within 30 days.

That’s a participation right, not a veto. But a ChatGPT pilot that summarizes patient charts is exactly the technology Article 42 describes. So here’s the question the public record leaves open: did UCSF’s Professional Practice Committee help select and validate this pilot? UCSF’s CEO said in OpenAI’s announcement that the system is “engaging frontline teams to validate these capabilities.” He didn’t say whether nurses, or their committee, are among them.

Nurse.org reached out to OpenAI and UCSF Health for comment and will update this article with any responses. 

At HCA, the protection is narrower. Contracts ratified in October 2024 cover about 8,500 nurses at 17 unionized HCA facilities in six states, not the company’s roughly 180 hospitals. The union says the language ensures nurses have a say in new technology like AI so it “enhances, not diminishes, patient care.”

The UC language isn’t a one-off. Nearly identical wording appears in another CNA contract in UC Berkeley’s inventory of technology bargaining language. The union is working from a template, and nurses have now won AI language at at least seven health systems we track. We covered how in our report on nurses bargaining over AI.

The bigger picture: 94 of the 100 systems in our Nursing AI Watch database run Epic. If this spreads past the pilot, it lands at hospitals where no contract language may exist at all. We’ve seen what that looks like: at Montefiore, AI-adjacent restructuring reached utilization review nurses before most of them knew a change was coming.

Much of the online reaction focused on the business side, with investors and founders declaring that OpenAI plus Epic just killed the health-tech AI startup.

The reality, at least today, seems more limited. This integration is read-only, so it doesn’t currently replace the AI scribes and documentation tools many nurses already use, though OpenAI and Epic could both push deeper into those workflows from here. Epic sells its own chart-summarizing AI. And OpenAI calls Epic a “supported EHR,” not an exclusive partner, so every hospital still decides what to switch on.

One category is genuinely squeezed: tools whose whole job was summarizing the chart. For nurses, that shakeout has one practical meaning. The AI tool your hospital trained you on this year could be replaced, absorbed, or shut down as the market consolidates, and the retraining lands on you. Our database tracks more than $4.8 billion in disclosed vendor funding across the systems we follow. We’ll be watching which tools survive.

The bigger picture here is that when health tools leave nurses out of the conversation, it not only does a disservice to patients, but it devalues the entire nursing profession. Physicians are a part of a patient’s healthcare team, not the sole endpoint. If a ChatGPT plug-in will be used in a hospital or facility setting where nurses are part of the documentation and care, nurses should be part of the conversation and the development.

Nurses discussing the news publicly have raised sharper evaluation questions than most vendor checklists. Bring them to your unit council, practice committee, or shared governance meeting:

• Does the tool catch context that lives in nursing notes, or only in physician documentation?
• Can a confident summary create false reassurance about a patient who’s declining?
• Does it change how fast a nurse escalates a concern?
• Does it actually save time, or just move the work somewhere else?

Those are nursing questions. On the evidence from OpenAI’s own announcements, nurses haven’t been asked to answer them yet.

Related Nursing AI Watch Analysis:

🤔Has your hospital asked nurses to help evaluate AI tools before rolling them out? Share your experience in the comments below.