Inside Sheba Medical Center’s Push to Build an AI-Powered Hospital

TLDR

• Israel’s Sheba Medical Center is building a fully AI-powered hospital, where AI is embedded into clinical workflows while physicians remain responsible for medical decisions.
• The hospital, which Newsweek named as the seventh best in the world and no. 1 in the Middle East, is OpenAI’s first international hospital partner.
• Sheba has developed its own AI system, which it will pair with ChatGPT for Healthcare to develop what it calls “super doctors.”

Sheba Medical Center – which Newsweek ranked as the 7th best hospital in the world in 2026 – is working toward becoming a fully AI-powered hospital.

The Israeli medical center, the largest in the country and one of the biggest in the Middle East, has spent a little more than a year building its own AI platform for doctors and is now piloting ChatGPT for Healthcare, giving clinicians access to medical research and the hospital’s own protocols at the point of care.

Sheba is OpenAI’s first international hospital partner.

The hospital’s AI system helps reduce the administrative load of clinicians while ChatGPT for Healthcare provides access to the latest published medical research to help doctors make better decisions. Longer term, Sheba plans to create a connected system where AI agents perform much of the hospital workflow – all while doctors remain responsible for medical decisions.

“We don’t see a future when there are no humans. We see a future where there are superhumans,” Alon Agmon, director of R&D at Sheba’s AI Center, said in an interview with The AI Innovator. “What makes humans superhumans or ‘super doctors’ is AI. This is what AI can do for us. It can make their work much more informed, efficient and accurate.”

Sheba’s approach offers a glimpse of what enterprise AI adoption may look like once organizations move beyond giving employees chatbots and begin redesigning workflows around the technology.

While other health care providers have also rolled out AI tools, Sheba’s broader ambition is its focus on agent orchestration that can move from conversation to recommendation to physician authorization to action.

Starting with the emergency room

Sheba has invested in technology, startups and partnerships for years. But as generative AI emerged, hospital management concluded that adopting the technology would require something different: an internal organization capable of changing the hospital from within.

Management also learned they needed to be serious about investing in AI because it wasn’t a cheap effort.

“This is the first thing we said at the outset to management,” Agmon said. “We said, ‘listen, we can transform health care, we can transform this hospital, but it’s going to cost.’”

Sheba made the bet. Not only is it a large institution – it has 2,000 beds and treats over two million patients a year – it has an investment portfolio valued at $6.7 billion through ARC (Accelerate, Redesign, Collaborate), the innovation organization it founded in 2019 that backs health technology startups such as Aidoc.

Sheba created what Agmon describes as the startup-like AI Center, recruiting people from the technology industry – Agmon joined from Microsoft – and embedding clinicians in the organization. Some doctors work part time as product managers, helping engineers identify clinical problems where AI might be useful.

The team began in the emergency department. Doctors there spent considerable time documenting encounters, collecting test results and preparing discharge or admission letters. Sheba built an internal platform called SmartER to automate parts of that work.

A physician can record a conversation with a patient rather than type notes during the visit. The system transcribes and summarizes the conversation into the medical chart. As a patient moves through the emergency department, it can also incorporate information such as blood tests and other results into a running summary.

By the time the physician is ready to make a decision, much of the information has already been assembled – something that normally makes patients wait for hours in ER. The platform can also prepare discharge and admission letters and surface potential discrepancies in the record.

Demand for SmartER spread beyond the emergency department. It is now used in other areas of the hospital, including internal medicine and clinics, with services tailored to different departments.

The productivity twist

The obvious expectation was that automating documentation would also shorten the amount of time patients spent in the emergency department.

It didn’t. “Waiting time didn’t go down,” Agmon said. “But patient experience and physician experience went up.”

His hypothesis is that doctors and patients simply filled the newly available time. With less attention devoted to typing, patients asked more questions and doctors gathered more information. The result was more comprehensive documentation rather than substantially shorter visits.

The system also handles conversations in Hebrew, English, Arabic and Russian and can translate them into Hebrew for the medical record.

Sheba consequently measures AI’s return in several ways, including documentation quality, physician and patient experience, time spent on administrative work, the number of departments using AI and how many doctors use it daily.

Sheba is now adding another layer. The hospital began an early pilot of ChatGPT for Healthcare around the end of July and plans to expand availability more widely by the end of September, Agmon said.

How SmartER and ChatGPT fit together

For now, ChatGPT for Healthcare and SmartER are separate systems, although Sheba plans to explore integrating them.

The distinction reflects their different jobs. SmartER is primarily embedded in hospital workflows and patient documentation. ChatGPT for Healthcare is intended to give clinicians an AI assistant connected to medical literature and Sheba’s institutional knowledge.

Doctors can use it from a computer or phone to research questions that arise as they move between patients – such as how a medication works, appropriate dosage or what the medical literature says about a particular condition. It can also help researchers conduct literature reviews and physicians prepare for rounds.

Sheba can connect its own protocols, papers and institutional knowledge to the system so an answer can incorporate both external medical research and the hospital’s practices.

That is intended partly to address something already happening inside hospitals: Doctors are using general-purpose AI tools such as the consumer-version of ChatGPT, Gemini and other chatbots. “We wanted to give them a much more robust platform that will give them proper answers that we know are of the highest quality,” he said.

That also meant providing a controlled alternative with medical sources and hospital-specific guardrails.

How Sheba protects patient data

That control becomes particularly important when doctors are dealing with patient information.

Agmon said Sheba does not use patient data to teach ChatGPT its institutional knowledge. Instead, it connects information without patient context, such as hospital protocols and medical know-how.

The hospital has also built protections against accidental disclosure.

If a doctor pastes material from a medical chart into ChatGPT and inadvertently includes a patient’s name, address or other prohibited information, Sheba’s system can detect it before the information is sent to OpenAI. It can redact the identifying information or block the request and tell the physician to edit it. Similar protections apply to uploaded files and images.

Sheba also maintains a team focused on AI evaluations and guardrails. Systems are evaluated during development and after deployment, Agmon said, with monitoring designed to catch problems including hallucinations, contradictions and changes in the quality of responses.

Generated letters and summaries are also checked against information in the medical chart for conflicts or contradictions. “We always keep evaluating our systems,” he said.