Aerial view of the Sheba Medical Center campus in Tel HaShomer, Israel, which brings together acute-care, pediatric, rehabilitation, cancer and other specialty facilities. Credit: Sheba Medical Center
OpenAI’s first international hospital partnership is taking shape at a non-English-speaking medical center that does not use electronic health records (EHRs) from Epic, the prominent U.S. electronic health record vendor. That institution is Sheba Medical Center, the Israel-based medical center with its own 50-plus-person AI Center within ARC, its innovation arm.
The partnership with OpenAI is part of Sheba’s aim to become “AI-powered” by 2030. “We really want to democratize AI tools for the campus, for the people working in the hospital, whether they are clinicians or other professions,” said Ayelet Akselrod-Ballin, head and CTO of Sheba’s AI Center in an interview at Ai4 in Las Vegas.
The partnership originated inside Sheba, with its engineering leadership approaching OpenAI. “It’s something that came out from a real need and a real vision that to really build an AI hospital means not only to bring in the technology but also to educate the people and to give them the ability and sort of self-service, whatever technology is out there.”
The hospital says the agreement will provide clinicians across its network and on mobile devices with access to ChatGPT for Healthcare, while implementation across clinical, research and operational teams will begin “in the coming months.” The partnership also gives Sheba early access to OpenAI’s latest healthcare models before they become broadly available.
Ayelet Akselrod-Ballin
Akselrod-Ballin said the longer-term vision includes giving employees mobile access to OpenAI’s models along with the ability to build their own agents and use the tools for research. The potential applications extend beyond clinical care. Akselrod-Ballin also cited surgical, regulatory and financial workflows as candidates for the same approach.
Already, Sheba has several AI projects in deployment. SmartER is ambient “invisible” AI in the emergency department, built with the clinical documentation startup ScribeMD, that listens to the patient-physician conversation, transcribes it, and generates a structured clinical summary for physician review. Project K, developed with Microsoft Israel and KPMG, goes further into the workflow, running intake through an AI avatar and providing an automated, AI-reliant process. A third system, Beyonder, screens admissions and routes stable patients to Sheba Beyond, the virtual hospital, while flagging complex cases for in-person care.
Doing more with less
The rollout comes as Israel and much of the world deals with a shortage of clinicians. OECD’s 2025 profile reports 3.5 practicing doctors per 1,000 people in Israel, below the OECD average of 3.9. Japan, meanwhile, has 2.6 practicing doctors per 1,000 people, among the lower rates in the 38-member Organisation for Economic Co-operation and Development. The WHO European Region faces a projected shortage of nearly 1 million health workers by 2030, according to the World Health Organization, while the U.S. could face a shortage of up to 86,000 physicians by 2036, according to the Association of American Medical Colleges (AAMC).
Akselrod-Ballin frames the shortage as a “crisis in healthcare,” and AI can help bridge the gap by reducing the time clinicians spend on administrative work and searching for information. “We’re heading towards a situation where there are not enough clinicians to really support patients.”
Project K at Sheba Medical Center, an AI-enabled emergency-care environment designed to support automated intake and triage workflows. Credit: Sheba Medical Center
Akselrod-Ballin has spent years working on AI for radiology. In a 2019 Radiology study from her time at IBM Research, she and colleagues developed an AI model combining mammograms with electronic health-record data. In a retrospective test set, the model identified 34 of 71 false-negative mammograms.
Other studies have found operational gains elsewhere in radiology. A 2023 cluster-randomized trial at four stroke centers that used AI software to analyze CT angiograms for large-vessel occlusions and alert clinicians reduced the time from hospital arrival to thrombectomy initiation by 11.2 minutes. More recently, a prospective study of generative AI-assisted radiograph reporting found a 15.5% improvement in documentation efficiency across nearly 24,000 interpretations, with peer review finding no significant difference in clinical accuracy or report quality.
Despite gains in image interpretation and triage, Akselrod-Ballin sees generative AI as a distinct wave from earlier radiology AI. “You’re working at the basic level of reducing the waste of time involved in searching in these EMRs, summarizing all the text, doing all this administrative work,” she said. “That really does not require the knowledge that nurses, radiologists or other clinicians have.”
Validating a technology that is still emerging
While the OpenAI partnership expands Sheba’s roster of outside AI partners, the medical center also has substantial internal capacity to develop AI technologies through its ARC Center. Its 50-plus-person AI Center has worked on projects including SmartER, an ambient AI platform for emergency-room workflows, and Project K, which applies AI across intake, triage and other ER tasks. ARC’s broader startup ecosystem also includes AI-focused companies such as AISAP, whose FDA-cleared platform uses AI to support point-of-care ultrasound.
Sheba’s work on emerging AI workflows sometimes reaches areas where established guidelines remain limited. “There is some regulation, of course, but we are in a way building the regulation, the guardrails, the monitoring, the policy, the governance,” Akselrod-Ballin said. “We’re building it together as we go.” She described that process as one advantage of having an AI Center embedded inside the hospital, where its teams can work directly with clinicians, IT and security.
Alongside the OpenAI rollout, Sheba is trying to build AI capabilities that can extend across departments. The AI Center maps problems that recur across the medical center and looks for infrastructure and components it can reuse, adding specialized agents or new platforms where workflows diverge.
Akselrod-Ballin said part of the plan is to “map the big problems and see which ones are wide across the hospital.” She pointed to Sheba’s five emergency departments, including dedicated oncology and children’s ERs. “If you’re solving an ER workflow, at Sheba we have five different ERs,” she said. “What you build for the ER, most of it would be relevant for some of the departments.”
Building and scaling agentic workflows
Agentic workflows also open up new considerations. The aim is to create an agentic infrastructure whose components can carry over to new workflows. “We’re trying to really make a generalizable approach,” Akselrod-Ballin said. “You build an infrastructure that is agentic, where you have different agents working on different tasks, and then you reuse as much as you can components that you’ve already built.” A sufficiently different workflow could require a new agent or, in some cases, a separate platform.
ARC’s international hospital relationships could also provide a channel for sharing technology and implementation experience. “We are in close relationship with some of these hospitals, and we are happy to collaborate and share things that we built and share things that they’ve built and share knowledge of how things are done,” Akselrod-Ballin said.
The same workflow approach could extend into clinical research. Akselrod-Ballin pointed to the many operational and administrative tasks inside a medical center that could be handled partly by agents. “There are so many things inside the medical center, operative tasks and administrative tasks that you would like to make more efficient, and you can really train agents to do and build a really multi-agent or agentic platform that does all of this,” she said.
Clinical trials are one potential application. Akselrod-Ballin said companies already offer technology for some trial workflows, and Sheba would continue weighing whether to build or buy additional capabilities. She also pointed to patient selection and the broader process of breaking clinical-trial workflows into components that can be addressed across a platform.
“I’m sure that the clinical trials will be changing the way that they work and will make them significantly faster very soon,” she said. “So things will be moving very fast.”
Filed Under: clinical trials, Drug Discovery, machine learning and AI
Tagged With: agentic workflows, AI in healthcare, ambient AI, ARC Innovation Center, Ayelet Akselrod-Ballin, ChatGPT for Healthcare, clinical documentation, clinical trials, EMR optimization, generative AI, healthcare workforce shortage, hospital automation, OpenAI, radiology AI, Sheba Medical Center, SmartER