Sheba OpenAI hospital deployment ties AI answers to clinical rules

The Sheba OpenAI hospital deployment will give physicians, nurses, researchers and other staff secure access to an AI-powered clinical reasoning platform, Sheba Medical Center announced Tuesday. The Israeli hospital said the system will summarize peer-reviewed medical research and clinical guidelines, with citations included in responses, while a later phase will add Sheba’s own protocols and policy documents.

The deal is OpenAI’s first international hospital deployment, according to Sheba. Its significance lies less in replacing clinical judgment than in adapting a general AI system to the operating rules of a specific institution, a pattern already visible in banking, where AI tools have been built around compliance, risk controls and audit requirements.

How will Sheba use OpenAI in its hospital?

Sheba said clinicians and staff will use the platform to support clinical reasoning by drawing on medical literature and guidelines. The hospital plans to integrate its internal clinical protocols and policies so the system’s answers reflect Sheba’s own practices as well as broader medical evidence.

Dr. Eyal Zimlichman, Sheba’s chief innovation officer and founder of its ARC innovation arm, said the hospital is “not just implementing AI in medicine but building a fully AI-powered hospital.” i24NEWS reported that OpenAI’s models will not be trained on Sheba’s data and that patient information will be protected through data isolation and audit protocols.

Sheba said final medical decisions will remain with its clinicians. The hospital will also receive early access to OpenAI’s newest models through the company’s API for research purposes.

Why hospitals are adapting AI to local rules

A general-purpose clinical AI system can reflect published research and accepted medical guidance, but hospitals also rely on local treatment pathways, committee-approved protocols and operating policies. Adding those rules can make an AI tool more relevant to how a particular hospital delivers care, while still leaving accountability with licensed clinicians.

Financial institutions faced a similar issue earlier. PYMNTS reported that Anthropic launched 10 financial-services AI agents for tasks including underwriting reviews, know-your-customer checks and compliance work, with the tools designed to connect to banks’ existing risk and compliance systems. PYMNTS also reported that Experian built an Agent Operating System inside its Ascend Platform to give lenders auditability and human oversight across stages of AI-assisted lending decisions.

The comparison is institutional rather than clinical: banks needed AI systems that could account for their lending policies, fraud limits and regulatory obligations. Hospitals are now applying the same logic to medical workflows, where an answer based only on general literature may not match an individual institution’s approved care process.

Which other hospitals are using clinical AI this way?

Cedars-Sinai announced in May that it was working with OpenEvidence, a clinical AI platform, and planned to add its own care pathways and best practices so clinicians could view medical literature alongside Cedars-Sinai-specific guidance, Newswise reported. OpenEvidence CEO Daniel Nadler said, “Medicine is not practiced in the abstract. It is practiced on individual patients with unique histories and complexities.”

Mount Sinai adopted a related approach two months earlier, according to Becker’s Hospital Review, becoming OpenEvidence’s first enterprise health system customer. The seven-hospital network gave physicians, nurses and pharmacists access to a platform drawing on clinical guidelines and peer-reviewed literature. Girish Nadkarni, Mount Sinai’s chief AI officer, said the system gives care teams “real-time access to rigorously sourced, evidence-based insights.”

The deployments point to a shift in how hospitals are buying AI: the model is only one component. The differentiating layer is increasingly the institution-specific content, controls and review process that determine how AI-generated answers fit into clinical practice.

This story draws on original reporting from PYMNTS.