For years the patient chart has been the thing everybody complains about and nobody can get out of. It is where a doctor's notes, lab values, prescriptions and referrals live, and in the United States most of those records sit inside software made by a single company, Epic. On the September 5 episode of Moonshots with Peter Diamandis, the panel described Epic as holding records for 325 million patients — on the show's telling, roughly the entire U.S. population — and Peter Diamandis, who said he had been involved in that business through his health venture Fountain Life, said that from the inside it had been "a bear to navigate" for physicians, while "patients have never had access." His verdict on putting an AI layer on top of it: "awesome."
The news behind the remark is an OpenAI announcement on September 1: healthcare organizations can now connect authorized electronic health record data, and additional industry data, to ChatGPT.
What the integration actually does
An electronic health record, or EHR, is the software system that stores a patient's clinical history for a hospital or clinic. The new connection means a clinician working in ChatGPT for Healthcare can put questions to that record in ordinary language rather than clicking through screens.
OpenAI describes the kinds of questions the integration is meant for: what has changed since a patient's previous visit, what the laboratory results show, which medications were changed, and whether any referrals were ordered and never resolved. The answers come back as summaries that point back to the supporting information in the chart, so the person reading a summary can check it against the record it came from.
On the podcast the same capability was described from the clinician's side: pulling appointment notes, lab results and medications, and asking questions across a patient's entire record rather than one document at a time.
OpenAI's announcement covers two arrangements that are easy to confuse. In one, the record's context is brought into ChatGPT. In the other, ChatGPT is embedded inside the EHR workflow itself, so the clinician stays in the system they already use. Supported deployments can offer either.
There is also a third piece that is not about patient data at all. A separate Healthcare Public Data integration gives structured access to public sources, among them PubMed for published medical literature, DailyMed for drug labeling information and CMS Coverage for U.S. insurance coverage rules. That is reference material, not anyone's chart.
UCSF Health is named as a pilot partner, validating practical uses with frontline teams. The deployments are organizational and authorized ones, with a health system testing what the tool is good for in daily work.
Separately from all of this, and aimed at ordinary people rather than institutions, the show noted that consumers can connect services such as Apple Health, One Medical and Function Health so ChatGPT can help them understand test results or prepare for an appointment. That is a different product path from the organizational integration, and it does not involve a hospital's authorized record system.
From a records connection to a proposed standard of care
The panel did not linger long on the plumbing. Emad Mostaque took the Epic integration as the starting point for something much larger: a sprint, he said, so that "within a year or two max, every single health decision is double-checked by an AI."
Another panelist, speaking to Mostaque by name, went further. "I think it's going to become malpractice to diagnose a patient without AI in the loop," that panelist said, adding, "We already know AI is a far better physician, a diagnostician, than a human is." The same speaker sketched what enforcement might look like: a series of approved models, some running on local devices at the edge and some in the cloud, with the requirement that every diagnosis has had at least one AI check. "That will save so many lives. It will detect so many cancers."
The remarks then moved to the volume of information as much as diagnosis, with no change of speaker marked in the transcript. The data collected around cancer and other conditions today, the argument ran, is tiny compared with what could be gathered and processed with AI transforming it, and the goal was put plainly: "No one should have to die of cancer."
Both of these are proposals made on a podcast rather than descriptions of how medicine is practiced or regulated. OpenAI's announcement is about connecting authorized records and public datasets to a clinical assistant whose summaries cite the chart; it sets out no approval scheme for diagnostic models and makes no claim that a model outperforms physicians. The panelist offered the comparison of AI and human diagnosticians as a personal assessment, without citing a study. The malpractice line is that panelist's forecast about where standards will go, and the one-to-two-year sprint is Mostaque's target, not a program anyone has funded.
One of the panelists pointed to that gap. It struck that speaker as strange that, with programs such as Genesis already running and the capability apparently within reach, no one simply directs $10 billion at curing these diseases: "It's tractable."
The unglamorous part
What drew the panel's interest in the Epic connection was the friction it removes. A record that took clicks and training to navigate can now be asked a question. Whether that becomes a second opinion on every diagnosis, as the panel wants, depends on evidence, approvals and institutions that do not yet exist. What exists today is narrower: a clinician asking what changed since the last visit, and a summary that shows which part of the chart it came from.