Mayo Clinic Now Runs 150 AI Models Inside the Hospital

Walk into an exam room at Mayo Clinic in Rochester, Minnesota, and a good deal of the work has already happened before the patient sits down. Many people who come to the clinic are seeking a third or fourth opinion, and they arrive carrying disordered stacks of paperwork from other health systems — dozens, sometimes hundreds of pages per person. For years, physicians simply read through all of it. Now a growing layer of software reads first, and the doctor reads second. It is a small change in sequence, but it hints at a larger rearrangement of how medicine is practiced.

What Happened

In mid-July 2026, Mayo Clinic offered an unusually candid look at how deeply artificial intelligence has been woven into its daily operations. At the center of the story is a tool called Record Time, built with Scale AI, that parses incoming medical records, generates a relevant patient summary, arranges documents in chronological order, and makes the whole file searchable. Dr. Alexander Ryu, an internal medicine physician who also serves as vice chair of innovation for the Department of Medicine, said it saves him somewhere between five and thirty minutes of preparation per visit, depending on how complicated the case is.

That time is not trivial. Mayo Clinic receives tens of millions of pages of outside records every year, and the risk is not only slowness but oversight — a decisive detail buried on page ninety that changes a treatment plan. Record Time is designed to surface exactly those details. And it is far from the only such system. According to Dr. Matthew Callstrom, a radiologist and the medical director of Mayo Clinic's generative AI program, there are now roughly 150 AI models deployed across the hospital, developed through partnerships with firms including Microsoft and Scale AI.

What is striking is not any single model but the accumulation. A hospital does not usually announce that it is running one hundred and fifty of anything. The number signals that AI at Mayo has moved past the pilot stage — past the press-release demo — and into the unglamorous machinery of routine care, where tools are judged less by their novelty than by whether clinicians quietly keep using them.

Why It Matters

The reason medicine is fertile ground for this technology is almost mundane: much of clinical work is pattern recognition across large volumes of data, and that is precisely what these systems do well. Jason Droege, the chief executive of Scale AI, framed the appeal in terms of tedium rather than magic — the idea that software can absorb the repetitive parsing that specialists otherwise do by hand, freeing them to reach a diagnosis faster and treat more people. Callstrom traces his own conviction back to 2016, when he watched AI help radiologists catch subtle, early cancer warning signs in imaging that were easy for a tired human eye to miss.

The clearest illustration of the stakes is a clinical trial now underway to see whether AI can flag patients at risk of, or already carrying, early-stage pancreatic cancer. Pancreatic cancer is one of medicine's cruelest diagnoses precisely because it is usually caught late; by the time it has spread regionally or metastasized, the five-year survival rate hovers around nine percent. If a model can read the faint signatures of the disease years before a physician would otherwise suspect it, the practical meaning is measured in survivable cases rather than in accuracy percentages. Mayo has also used AI to read patients' heart rhythms and estimate who might go on to develop atrial fibrillation, a condition that can quietly set the stage for clots and strokes.

These are not chatbots answering health questions. They are narrow, tightly scoped instruments pointed at specific, high-consequence problems — the kind of work where being early is the entire benefit, and where a few months of warning can change the arc of a life.

The Reaction

Inside the hospital, the mood is less breathless than the technology's boosters might expect, and that skepticism appears to be a feature rather than a bug. Callstrom describes physicians as instinctively doubtful, and Mayo leans into it: a new tool is offered, not imposed, and doctors are free to abandon it if it does not earn its place. The truest measure of a model's worth, he suggests, is its adoption rate — whether busy clinicians choose to keep it once the novelty wears off.

Not every reaction has been about clinical performance. Enthusiasm inside any large institution rarely arrives without friction, and Mayo's expansion has drawn scrutiny over privacy and oversight — a reminder that patient records are among the most sensitive data anywhere, and that the question of who governs their use is unsettled. The hospital says it is committed to responsible development, with privacy, security, and transparency embedded in its processes. There is also the perennial anxiety about jobs. So far, Callstrom says, roles are changing rather than disappearing: a nursing team helped build a system that listens during visits and drafts the notes, potentially halving the hour-plus that nurses otherwise spend each day typing — time redirected toward actually talking with patients.

What Comes Next

Mayo's method for getting there is deliberately slow. New AI tools move through something resembling a clinical trial: first a small group of patients under close physician oversight, then measured performance, then a wider rollout, then continuous monitoring once the tool is live. It is a governance structure borrowed from how medicine already evaluates drugs, and it stands in quiet contrast to the ship-fast culture of much of the software industry.

Droege, whose company builds these systems, is notably restrained about the pace. He calls the sweeping predictions that everything in health care will be fixed within a year or two "wildly ambitious," and argues that speed should not be the top priority. "Quality of care is the bar," he said, with speed coming second. It is an unusual thing to hear from a vendor, and it captures the particular tension of AI in medicine: the same institution must be both eager enough to deploy at scale and cautious enough to distrust its own tools.

Closing Thoughts

There is a temptation to read a story like this as either salvation or threat, and it comfortably refuses both. What is happening at Mayo is more modest and, in a way, more interesting: AI is settling in as a kind of first reader — the system that sorts the file, notices the anomaly, drafts the note — while the human remains the second reader who decides what any of it means. The doctor still meets the patient. The judgment still belongs to a person.

If there is a lesson in the 150 models quietly running behind the scenes, it may be that the most consequential version of this technology is not the one that promises to cure disease in a headline, but the one that gives a physician back thirty minutes and a slightly clearer view of the evidence. That is not a revolution announced with fanfare. It is a change you would only notice as a patient if you realized, halfway through the visit, that your doctor was looking at you instead of at a stack of paper.

한글 요약

미국의 대표적 병원인 메이오 클리닉(Mayo Clinic)이 2026년 7월 중순, 자사 진료 현장에 인공지능이 얼마나 깊숙이 들어와 있는지를 이례적으로 공개했습니다. 핵심은 스케일 AI(Scale AI)와 함께 만든 '레코드 타임(Record Time)'이라는 도구로, 외부에서 밀려드는 수천 페이지의 흩어진 진료 기록을 자동으로 요약·시간순 정렬하고 검색 가능하게 만들어 의사가 환자 한 명당 준비 시간을 5~30분 아낄 수 있게 합니다. 병원 내 생성형 AI 책임자인 매튜 콜스트롬(Matthew Callstrom) 박사에 따르면, 현재 병원에는 약 150개의 AI 모델이 실제로 배치·운영되고 있습니다.

의미가 큰 이유는 의료의 상당 부분이 방대한 데이터 속 패턴 인식이며, 그것이 바로 이 기술이 잘하는 일이기 때문입니다. 대표적으로 메이오는 통상 늦게 발견돼 국소 전이·전이 단계 5년 생존율이 약 9%에 그치는 췌장암을 수년 앞서 조기에 식별할 수 있는지 임상시험을 진행 중이며, 심장 리듬을 분석해 심방세동 위험군을 예측하는 데에도 AI를 활용해 왔습니다. 다만 병원 내부의 반응은 신중합니다. 의료진은 새 도구를 강요받지 않고 직접 써 본 뒤 채택 여부를 정하며, 콜스트롬 박사는 모델의 진짜 가치를 '채택률'로 판단한다고 말합니다. 환자 데이터의 민감성 탓에 프라이버시·감독을 둘러싼 논쟁도 함께 불거졌습니다.

메이오의 접근은 의도적으로 느립니다. 신규 AI 도구는 소규모 환자군·의사 감독 → 성능 측정 → 확대 → 지속 모니터링이라는, 신약 평가와 닮은 절차를 거칩니다. 스케일 AI의 최고경영자 제이슨 드로지(Jason Droege)조차 1~2년 안에 모든 것이 해결된다는 예측을 "지나치게 야심적"이라 부르며 속도보다 진료의 질이 우선이라고 강조합니다. 결국 이 이야기는 구원도 위협도 아닙니다. AI는 파일을 정리하고 이상 신호를 짚어 주는 '첫 번째 독자'로 자리 잡고, 무엇을 의미하는지 판단하는 '두 번째 독자'는 여전히 사람인 의사입니다. 가장 값진 변화는 질병을 단번에 정복하겠다는 헤드라인이 아니라, 의사에게 30분과 조금 더 또렷한 근거를 돌려주는 조용한 진전일지도 모릅니다.

참고: CNN, Mayo Clinic