Ukraine Will Choose AI Models No Foreign Provider Can Control

Ukraine's Ministry of Digital Transformation has made its position clear: when choosing artificial intelligence systems for government agencies, businesses and the military, one criterion will matter above all others - whether the model can run on the country's own servers, without any possibility of a provider remotely restricting or shutting it down.

The announcement came against the backdrop of a war in which Ukraine has spent years testing just how much of the state's critical digital infrastructure can safely depend on foreign technology providers. Kysliy stressed that the issue isn't where a model comes from, but who actually controls it.

This confirms that AI sovereignty isn't just a defense slogan, it's a necessity - Roman Kysliy, Chief AI Officer, Ukraine's Ministry of Digital Transformation

What Triggered the Decision

The immediate trigger was reportedly a directive from the US government ordering Anthropic to restrict access to some of its most powerful models. For Kyiv, that was a signal that even allied relationships offer no guarantee of continued access to tools that government, business and the armed forces are increasingly coming to rely on.

Kysliy put it bluntly: in this view, a language model is a commodity, not an ally. What matters is whether the provider will agree to let its software run on infrastructure controlled by Ukraine, free of remote restrictions, feature-disabling updates or outside audits.

A Homegrown Model Built on Gemma

The government project, run jointly with telecom operator Kyivstar, is built on Google's open Gemma model family. The ministry had previously tested Mistral and OpenAI's open GPT-OSS as well, but the Gemma variant matched the performance of remote alternatives in benchmarks while still being deployable locally.

Training is taking place on Google's computing infrastructure, since Ukraine currently lacks sufficient capacity of its own. Eventually, though, the model is meant to move onto servers inside the country, so no foreign entity has any influence over its operation. The system is expected to go live in autumn 2026, covering public administration, business and military applications.

Today, the flagship government app Diia, used by millions of Ukrainians for official business, runs on Google's Gemini model accessed remotely through servers in the European Union, partly under a free token allotment. That is meant as a stopgap until a fully sovereign alternative exists.

What It Means for Other Countries

Kyiv's decision fits a broader trend of governments starting to treat control over AI models as a matter of national security, alongside control over energy or communications. Ukraine's war has long served as a testing ground for new technology, from drones to image-recognition systems, and AI sovereignty now appears to be joining that list.

For smaller European countries, including Poland, the signal is clear. Making critical public services dependent on models available only through a foreign provider's cloud carries political risk, not just a technical one. Projects like Poland's Bielik model or European initiatives for sovereign language models gain extra justification in this light.

It remains an open question whether a Gemma-based model can truly match the capabilities of closed commercial systems in day-to-day use by government and the military, where the stakes are often not convenience but information security during an active conflict.

Sources: Ukraine to pick AI models operated without provider control, official says (investing.com), Reuters via investing.com, Ukraine Moves Toward a Sovereign AI Model (digitalstate.gov.ua)