What Happened
On September 22, a group of North Carolina health systems and universities launched the North Carolina Collaborative Health AI Network, or NC CHAIN. The Duke Endowment is backing it with a three-year, $4.4 million grant. The goal is narrow and practical. Rural hospitals, critical access hospitals and small clinics across the state should be able to learn about, choose and safely run AI tools without building an entire data science department first.
The network is housed at UNC-Chapel Hill, but it is not a single-campus project. According to the university's announcement, the founding partners include UNC-Chapel Hill, Duke University, UNC Health, Duke Health, the Cecil G. Sheps Center for Health Services Research and the Duke-Margolis Institute, along with several other state organizations. Two health systems that usually compete for patients, staff and research money are putting their governance playbooks into a shared pool.
Dr. David McSwain, UNC Health's chief medical informatics officer, is the principal investigator. Dr. Armando Bedoya, chief data and analytics officer at Duke Health, is a co-investigator. Saif Khairat, UNC-Chapel Hill's chief AI officer and vice provost for AI, is also involved from the university side. Their pitch is not that small hospitals need their own models. It is that they need a way to judge the tools vendors are already trying to sell them.
The mechanism is sharing. Instead of every rural hospital writing its own AI policy and vetting every vendor from scratch, NC CHAIN plans to pass around frameworks, vendor evaluations, lessons learned and direct technical help. Community health organizations are supposed to help decide what the network works on first. The organizers also say the model is built to add more North Carolina health systems and, later, to be copied in other states.
Why It Matters
The clearest explanation came from Bedoya. Big systems like Duke and UNC have data scientists, compliance officers and informatics staff who can evaluate and monitor an AI tool. "Most rural hospitals and small practices do not," he said. That one sentence is the whole problem. An algorithm that flags sepsis risk, drafts clinical notes or predicts no-shows is only as safe as the people checking whether it works on their own patients.
National data shows the gap is already large. A study by Oklahoma State University economist Brian Whitacre, reported by StateImpact Oklahoma in August, used the American Hospital Association's 2023 IT survey of more than 2,300 hospitals. It found that 81 percent of urban hospitals used predictive AI, compared with 51 percent of rural ones. Critical access hospitals were less likely to adopt it. Hospitals in larger systems and with stronger finances were more likely to.
The type of use differs as well. Whitacre found rural hospitals were about as likely as urban ones to use AI for billing and operations, but less likely to use it for clinical decisions or treatment recommendations. They were also almost twice as likely to rely on models supplied by their electronic health record vendor instead of building or buying their own. In practice, that means many small hospitals are using AI that someone else chose, tested and tuned, often with little ability to check how it performs locally.
This matters in North Carolina because the rural footprint is large. The state's rural health programs describe nearly 3 million rural residents across 85 of its 100 counties. The state's Office of Rural Health supports 20 critical access hospitals and 12 other small rural hospitals. For many of those facilities, one IT staffer may be responsible for the electronic record, the network and every new software contract. Asking that person to also audit an AI model for bias and drift is not realistic without outside help.
Reaction
The launch fits a wider shift in how hospitals are being asked to handle AI. The Joint Commission, the main U.S. hospital accreditor, introduced a voluntary Responsible Use of AI in Healthcare certification on June 1. Hackensack Meridian Health in New Jersey became the first system to earn it in July, after a review of its governance, privacy safeguards, bias checks, monitoring, patient transparency and staff training, as Healthcare Dive reported. Those are exactly the areas a small rural hospital has the hardest time staffing.
McSwain framed the risk directly. He called the growth of the digital divide "a very real and poorly addressed risk" of fast AI adoption in healthcare. Brian Cass of the Sheps Center said the network does not intend to control how tools are deployed, and he described it as something with no close equivalent anywhere in the country. That is a big claim, and the network will have to prove it with results. But the design choice is notable. It treats rural providers as members who shape the agenda, not as customers receiving a finished product.
Researchers outside the state have pointed to the same need. Whitacre has argued for a kind of "HITECH 2.0," comparing today's situation to the 2009 federal program that spent about $27 billion pushing hospitals onto electronic health records. Rural hospitals lagged behind then as well. NC CHAIN is much smaller than a federal program, but it tests the same idea: early support for slow adopters may be cheaper than trying to close the gap later.
What's Next
The immediate work is building the network itself. That means recruiting rural and community members, setting up the shared resources and deciding which problems to take on first. The UNC Health announcement describes the goals as helping rural providers safely adopt AI tools, strengthen their finances and improve patient care. The finance part is important. Ambient documentation tools and billing assistants are often the first AI products small hospitals buy, because they can reduce claim denials and paperwork almost immediately.
The timing also lines up with new federal money. North Carolina received $213 million for 2026 through the federal Rural Health Transformation Program and could receive about $1 billion over five years, according to the state health department. Technology is one of the areas that funding is meant to support. A network that already knows how to evaluate AI vendors could help rural facilities spend that money on tools that work, instead of on whatever product arrives with the best sales pitch.
The harder test will come later. Three years is not long, and the grant pays for infrastructure, not for the AI tools themselves. The network will need to show measurable results: fewer failed deployments, faster and safer vendor reviews, and clinicians at small hospitals who trust the tools they use. If it can show that, the model could spread to other states. If it cannot, it may end up as another well-meant pilot.
Closing Thoughts
Much of the AI conversation in medicine focuses on the model: its accuracy, its benchmark scores, its latest version. NC CHAIN focuses on the less visible work around the model. Someone has to decide whether a tool fits a hospital's patients, watch whether it keeps working and notice when it quietly starts failing. Large academic systems pay whole teams to do that. Small hospitals mostly cannot.
That is why this story matters beyond North Carolina. The risk with AI in healthcare is not only that some hospitals will lack it. It is that they will get it anyway, bundled into software updates and vendor contracts, without the capacity to question it. A patient in a small mountain town deserves the same scrutiny of the algorithm that a patient at a large medical center gets. Sharing governance expertise across a state is not glamorous, but it may be one of the more realistic ways to deliver that.
The $4.4 million grant is small next to what hospitals spend on software every year. Its value is in who it reaches and what it teaches. If the network works, the most important result will not be a new AI tool. It will be hundreds of small providers who can ask the right questions before they switch one on.
한글 요약
9월 22일 미국 노스캐롤라이나주에서 농촌·소규모 병원의 AI 도입을 돕는 주 단위 협력망 'NC CHAIN(North Carolina Collaborative Health AI Network)'이 출범했다. 듀크 재단(The Duke Endowment)이 3년간 440만 달러를 지원하며, 사무국은 UNC 채플힐에 둔다. UNC 헬스, 듀크 헬스, UNC 셰프스 보건서비스연구센터, 듀크-마골리스 연구소 등이 참여해 대형 병원들이 이미 만들어 둔 AI 거버넌스 체계와 벤더 평가 자료, 기술 지원을 농촌 병원과 필수의료 접근성 병원(critical access hospital), 소규모 의원과 공유하는 구조다.
배경에는 커지는 도농 격차가 있다. 오클라호마주립대 연구에 따르면 미국 도시 병원의 81%가 예측형 AI를 쓰는 반면 농촌 병원은 51%에 그쳤고, 농촌 병원은 임상 판단보다 청구·행정 업무에 AI를 쓰는 경향이 강했다. 전자의무기록 업체가 제공하는 모델에 의존하는 비율도 도시 병원의 약 두 배였다. 듀크 헬스의 아르만도 베도야 박사는 대형 병원에는 데이터 과학자와 컴플라이언스 인력이 있지만 대부분의 농촌 병원에는 없다고 짚었다. 노스캐롤라이나에는 필수의료 접근성 병원 20곳과 소규모 농촌 병원 12곳이 있다.
NC CHAIN은 향후 노스캐롤라이나의 다른 의료 시스템을 추가하고, 다른 주로 모델을 확산하는 것을 목표로 한다. 노스캐롤라이나는 연방 농촌보건혁신프로그램으로 2026년 2억 1,300만 달러를 받았고, 5년간 최대 약 10억 달러를 받을 수 있어 이 자금을 검증된 AI 도구에 쓰는 데 이 협력망이 역할을 할 수 있다. 다만 3년이라는 기간 안에 실패한 도입 감소, 더 빠르고 안전한 벤더 검토 같은 측정 가능한 성과를 보여줘야 한다는 과제가 남아 있다.