Chile creates Latin America's first public AI sandbox for healthcare: validate before you trust

By Zendoric · October 7, 2026.

Chile will have the first public artificial intelligence bank for detecting breast cancer within the state health system. The project, named the Sandbox Nacional de Inteligencia Artificial en Salud, was presented on Wednesday at the Summit para la Salud del Futuro and grows out of an alliance between public and private institutions. Its raw material: 166,000 mammograms from 75,000 Chilean women, gathered so that researchers, universities and companies can test their algorithms there before bringing them anywhere near a real patient.

The logic is that of a regulatory "sandbox": a controlled environment where a technology is tried out under supervision before being released into the real world, a concept that already exists in banking and fintech and that is now being transferred to health care. The platform draws on data from Hospital Digital, the service that provides diagnostic support to 27 of the country's 29 Health Services, and will make it possible to measure not just whether an algorithm detects tumors, but also where it fails and with what kinds of errors. Health Minister May Chomali described it as "the first sandbox exclusively for the health sector in Latin America". According to Corfo, the state economic development agency taking part in the initiative, the results of each algorithm evaluated will be published in open reports, and the images will remain in state custody and free of any data identifying patients.

The dates are worth reading carefully: the bank will only become fully operational in the second half of 2027, that is, almost a year after the announcement. In between will come the open calls and, predictably, the adjustments typical of any public infrastructure handling sensitive health data. That is no small detail in a field where announcement and actual rollout are usually separated by years.

As sector context, breast cancer detection is one of the areas where diagnostic AI has already shown measurable results outside the lab: the Swedish MASAI trial, published in The Lancet Oncology in 2023, found that mammography screening with AI support maintained a detection rate comparable to double human reading while markedly easing radiologists' workload. That is exactly the kind of evidence a sandbox like Chile's aims to generate systematically and publicly, rather than letting each vendor present its own accuracy figures with no one auditing them independently.

And that is what really matters about this announcement, beyond the headline: it is not a diagnostic tool, it is governance infrastructure. Any country can buy an algorithm from a company promising to detect the vast majority of tumors; few have a public, neutral mechanism with real data from their own health system to check whether that is true before putting it to work with patients. That it is the state — and not each hospital or each vendor separately — that holds the data and publishes the results changes the incentives: a company no longer competes only to sell the flashiest promise, but to pass a test that is made public regardless of the outcome.

This connects with something we repeat at Zendoric: the biggest change AI can bring to health care is not a smarter model, but the real possibility of delivering early, cheap diagnosis to systems with a chronic shortage of specialists, as is the case across much of Latin America. If the algorithms that pass this filter deliver on what they promise, the benefit is not only technological: it is radiologist time freed up for the hard cases and earlier detection for public system patients. That is, in miniature, the kind of health care abundance we are betting on over the long term: not a magic cure, but systems that get more out of every hour of a scarce specialist.

The short-term risk does not disappear by decree either. Validating an algorithm with 166,000 studies is a good starting point, but it is a small fraction compared with the volumes the big commercial models are trained on, and nothing guarantees that the open calls will attract the sector's most serious players, or that the sandbox will not end up, in practice, serving as a convenient seal of approval for whoever already intended to sell their product in Chile. The promise of transparency — open reports, state custody, anonymized data — is only worth something if it is honored call after call, not just at launch. That said, for a country in the region to move ahead and build this piece of infrastructure, instead of simply buying foreign certifications, is a sensible bet: before entrusting a diagnosis to a machine, you have to be able to demonstrate, with your own data and in public, when it gets it right and when it does not.