Hospital Italiano unveils TANA: an in-house health AI that bets on sovereignty over clinical data

By ElDebate.com.ar · October 25, 2026.

Hospital Italiano unveiled TANA, its in-house generative artificial intelligence ecosystem, at the 2026 Congress on Health Informatics, Innovation and Research. According to the institutional presentation, it is a copilot for healthcare teams. It aims to ease the administrative burden and the fragmentation of clinical information.

The tool lets users ask natural-language questions about clinical progress notes, diagnoses and medical histories. It also makes it easier to review medical records and integrate data. The stated goal is for professionals to spend more time on direct care and on their relationship with patients.

The project rests on three pillars, which Daniel Luna, head of the Department of Health Informatics, links to one central question. The first is institutional memory: ensuring that accumulated knowledge does not depend solely on the people on the teams. The second is data governance: privacy, security and traceability. The mechanisms cited include anonymization and architectures in which models query data that stays under the institution's control. The third is reducing bias in models trained mostly on data from the United States, Europe or China, which may fit poorly with the clinical, epidemiological, cultural and linguistic realities of Argentina and Latin America.

The source provides no data on performance, deployment or clinical validation. We do not know which model it runs on, how many professionals use it or what results the hospital has measured. TANA should therefore be seen as a statement of intent and architecture, not a proven capability.

Our take. The most valuable part of the announcement is the question Luna raises: who controls the knowledge, and who benefits from its use? Hospitals generally hold decades of medical records, and that archive is a strategic asset. Handing it over to an external platform without control means giving up power. A model that queries the data “in-house,” instead of taking it elsewhere, is a sensible way to gain usefulness without giving up sovereignty.

The bias argument is also well grounded. As a matter of industry context, a tool calibrated to the local population, prevalent diseases and language should be more reliable where it is used. But that is a hypothesis to be tested, not a result. Until evaluations are published, the advantage of a home-grown tool remains a promise.

The case fits an argument we have made before: in clinical AI, the value lies in allocating human work well, not in replacing it. Cutting hours of paperwork and searching through medical records is a modest use, but a high-impact one. And it is the kind of improvement that, over time, frees doctors and nurses for what no machine does well: being there for patients.

In the short term, the risks are real. A copilot that summarizes or answers questions about medical records can make mistakes, and a plausible error is more dangerous than an obvious one. It requires auditing, traceability and human oversight. In the long term, it is good news that hospitals in the region are building their own capabilities: it brings more personalized medicine closer and makes it less dependent on a handful of vendors. It is a small step toward a healthcare system with more time for each patient and more shared knowledge.