--- title: "Sheba Is Rolling Out ChatGPT. The Data Layer Decides." slug: "sheba-medical-center-chatgpt-healthcare-data-layer" author: "Mikel Amigot" date: "2026-09-13 14:00:00" category: "Premium" topics: "healthcare AI, Sheba Medical Center, ChatGPT for Healthcare, health system integration, clinical data, interoperability, healthcare CIO" summary: "Sheba will be OpenAI's first international hospital partner for ChatGPT for Healthcare, announced July 28, 2026. The constraint is underneath: symplr's 2024 survey puts 51% of health systems above 50 software solutions." banner: "" thumbnail: "" linkedin: | Sheba Medical Center announced on July 28, 2026 that it will become OpenAI's first international hospital partner, deploying ChatGPT for Healthcare hospital-wide across clinical, research and operational teams. Worth getting the shape right before it becomes a board slide. What was announced is an assistant that reasons over peer-reviewed literature, clinical guidelines and Sheba's own protocols, care pathways and policy documents, with citations. It is not a connection to the patient record — and Sheba's EHR is not Epic, so the Epic connector OpenAI shipped in September does not apply there. Which puts the real question one layer down. A hospital does not run five to fifteen systems. In symplr's 2024 Compass Survey of more than 280 respondents, 51% of hospitals and health systems reported more than 50 different software solutions in use; in the 2023 edition, 55% of organizations relied on 50 or more point solutions; in the 2022 edition, 24% reported between 151 and 500. That is the work no model does for you: → Reading across systems of record in place, rather than extracting into another copy that becomes its own compliance surface → Identity resolution across systems that never agreed on an identifier → Provenance on every retrieved fact — which system, as of when → Role-scoped access and audit trails enforced beneath the model, not requested of it And the ownership question that follows: when a vendor builds your integration layer, you have rented the most durable asset in your stack. With ibl.ai you own all the code and the data — self-hosted inside your own perimeter, model-agnostic across any LLM, usage-based with no per-seat pricing, deployable anywhere from your own cloud to a fully air-gapped network. #iblai #AgenticAI #EnterpriseAI #HealthcareAI #HealthIT #Interoperability --- ## The Short Answer **Sheba Medical Center announced on July 28, 2026 that it will become OpenAI's first international hospital partner, running ChatGPT for Healthcare hospital-wide in the coming months. It reasons over literature and Sheba's own protocols, not the patient record. The constraint underneath is the data layer: symplr's 2024 survey puts 51% of hospitals and health systems above 50 software solutions. With ibl.ai you own all the code and the data.** A hospital that has not solved its integration problem has not solved it by buying a better model. ## What did Sheba Medical Center and OpenAI actually announce, and when? A hospital-wide deployment of ChatGPT for Healthcare, [announced on July 28, 2026](https://www.einpresswire.com/article/929916693/sheba-medical-center-collaborates-with-openai-on-its-first-international-healthcare-deployment) — six weeks before this post, not this week. Sheba, in Ramat Gan, becomes OpenAI's **first international hospital partner** for the platform. Physicians, nurses, researchers and operational staff get access across the hospital network and on mobile devices, with rollout beginning "in the coming months." Two details in the announcement do most of the work. The platform synthesizes peer-reviewed research, clinical guidelines and public health sources **with citations and publication dates**, and Sheba is loading its own clinical protocols, care pathways and policy documents into it so answers reflect the institution's approved standards. And [OpenAI's models will not be trained on Sheba's data](https://hitconsultant.net/2026/07/28/sheba-medical-center-openai-chatgpt-for-healthcare-deployment/), with zero-data-retention and non-training terms. What the announcement does not describe is a connection to the patient record. It is an assistant over literature and institutional knowledge — a genuinely useful thing, and a different thing from an agent that reads a chart. Implementation runs through Sheba's ARC center, led by Prof. Eyal Zimlichman and Alon Agmon. And [Sheba's EHR is not Epic](https://www.drugdiscoverytrends.com/how-sheba-medical-center-became-openais-first-international-hospital-partner/), so [the Epic connector OpenAI shipped on September 1](/blog/openai-chatgpt-epic-ehr-platform-era-healthcare) does not reach it. ## How many disconnected systems does a hospital actually run? Far more than the five-to-fifteen figure that circulates. The published numbers are an order of magnitude higher. In [symplr's 2024 Compass Survey](https://www.symplr.com/press-releases/symplr-2024-compass-survey-shows-how-lack-of-alignment-impacts-us-healthcare-organizations), fielded among more than 280 respondents and published 8 October 2024, **51% of hospitals and health systems reported more than 50 different software solutions in use**. The [2023 edition](https://www.symplr.com/press-releases/2023-symplr-compass-survey-health-systems-highlights-urgent-need-align-cio-clinician-priorities-improve-healthcare-operations), of **210 members of CHIME**, the health-IT executives' association, had put it at **55% of organizations relying on 50 or more point solutions** to run healthcare operations. The [2022 edition](https://www.symplr.com/press-releases/symplrs-2022-compass-survey-cios-reveals-operational-inefficiencies-technology-led-opportunities), covering 132 CIOs, broke it down further: 35% ran 51–150 solutions, and **24% ran between 151 and 500**. These are operations systems — workforce management, provider data, contracting and spend, facility access, quality, safety and compliance — sitting alongside the clinical systems, not counting them. The executives running them say so plainly. [88% of the 2022 CIO respondents agreed](https://www.hcinnovationgroup.com/clinical-it/article/21287480/survey-cios-stressed-in-managing-overabundance-of-systems-platforms) that working with disparate IT systems and applications complicates their job, and in 2023, **84% agreed clinicians could redirect substantial time back to patient care** if that software were brought together. That is the labor the brief-level version of this story is pointing at: a standing internal cost, paid in salaried headcount, to move data between systems that were never designed to agree. ## Why does a literature-and-protocol assistant leave the integration problem untouched? Because the hard part of clinical data is not retrieving a document. It is assembling a patient. A literature assistant queries a corpus that is already unified, already deduplicated, and already carries citations. Loading a hospital's protocol library into it is a content problem with a known shape, and Sheba is solving it the sensible way. A chart is the opposite. The same patient exists under different identifiers in the EHR, the lab system, the imaging archive, the pharmacy system and the scheduling system, and reconciling them is the work that no model performs. This is where the [unified patient ontology](/blog/healthcare-ai-patient-data-ontology) problem starts, and it does not get easier as the model improves. It also does not get easier by extracting everything into one warehouse. In healthcare every copy of protected health information is a new surface to secure, audit, retain and destroy, and a copy goes stale in a domain where [staleness is clinical risk](/blog/healthcare-ai-bottleneck-is-data-not-models). The alternative is reading across systems of record in place, role-scoped and read-only, with the answer assembled at query time and provenance attached to every fact. ## Who owns the integration layer once an AI vendor builds it? Whoever holds the code. This is the ownership question most health systems are not asking yet, and it is the one that compounds. A model is a rental by design, and a good one: you switch providers when a better one ships. The integration layer is the opposite. It encodes how your institution's systems map to each other, which identifiers reconcile, which fields mean what, and which access rules apply to whom. That is institutional knowledge, accumulated over years, and it is the most durable asset in a hospital's AI stack. If a vendor builds it, the hospital has rented the durable part and bought the disposable part. Switching models later is easy. Switching out of someone else's connectors, identity resolution and access policy is a multi-year project. The same reasoning applies to the [multi-step clinical workflows](/blog/clinical-ai-workflows-need-pipelines-not-chat) built on top of it — prior authorization, coding, referral management — which encode the institution's own rules rather than the vendor's. ## What does a healthcare CIO need in the data layer before the model matters? Four capabilities, none of which arrives with a stronger model. - **Read-in-place access across systems of record**, role-scoped and read-only, so an agent assembles from live sources rather than from a copy that has to be secured separately. - **Identity resolution across systems that never agreed on an identifier** — the genuinely hard problem, and the one that determines whether cross-system context is possible at all. - **Provenance on every retrieved fact.** A clinician needs the source system and the as-of date. An answer without provenance cannot be safely acted on, however well cited the literature behind it is. - **Governance enforced beneath the model.** Role-based access bound to the existing identity provider, complete audit trails, and PHI handling enforced server-side rather than requested in a system prompt. Build these and any model becomes useful. Skip them and no model does. ## How does ibl.ai put the data layer inside a health system's own perimeter? By deploying the platform where the data already is, and leaving the integration layer in the institution's hands. With ibl.ai you own all the code and the data. The stack runs on the health system's own infrastructure with full source code access, is model-agnostic across any LLM so the institution can switch providers without rewriting the platform, is usage-based with no per-seat pricing, and can deploy anywhere — your own cloud, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity. Agents read EHR and ancillary systems [in place over HL7 FHIR](/solutions/medical-healthcare) under role-scoped permissions, with every access audited and access control bound to the institution's existing identity provider. The connectors, the identity resolution and the access policy are yours, which is what makes the model layer genuinely replaceable. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ibl.ai is family-owned and operated from New York, NY. *Related reading: [OpenAI wired ChatGPT into Epic — where does the PHI go?](/blog/openai-chatgpt-epic-ehr-platform-era-healthcare) — the companion question of where reasoning over protected health information runs once the chart is connected, and [healthcare AI's bottleneck was never the model](/blog/healthcare-ai-bottleneck-is-data-not-models) — why a partial record limits an agent more than a weaker model does.* *Sources: the July 28, 2026 announcement, the hospital-wide scope and the non-training terms from [Sheba's press release](https://www.einpresswire.com/article/929916693/sheba-medical-center-collaborates-with-openai-on-its-first-international-healthcare-deployment); the zero-data-retention terms and the ARC implementation leads from [HIT Consultant](https://hitconsultant.net/2026/07/28/sheba-medical-center-openai-chatgpt-for-healthcare-deployment/); Sheba's non-Epic EHR from [Drug Discovery & Development](https://www.drugdiscoverytrends.com/how-sheba-medical-center-became-openais-first-international-hospital-partner/); the partnership signing and clinician use from [Calcalist](https://www.calcalistech.com/ctechnews/article/rykzijsszx); the 51% figure from the [2024 symplr Compass Survey](https://www.symplr.com/press-releases/symplr-2024-compass-survey-shows-how-lack-of-alignment-impacts-us-healthcare-organizations); the 55% and 84% figures from the [2023 symplr Compass Survey](https://www.symplr.com/press-releases/2023-symplr-compass-survey-health-systems-highlights-urgent-need-align-cio-clinician-priorities-improve-healthcare-operations) of 210 CHIME members; the 151–500 breakdown from the [2022 symplr Compass Survey](https://www.symplr.com/press-releases/symplrs-2022-compass-survey-cios-reveals-operational-inefficiencies-technology-led-opportunities); the 88% figure via [Healthcare Innovation](https://www.hcinnovationgroup.com/clinical-it/article/21287480/survey-cios-stressed-in-managing-overabundance-of-systems-platforms).* ## Why does owning the AI stack matter? **ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.** - **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours. - **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform. - **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ. - **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.