July 8, 2026, (Inside AI) — Ukraine is prioritizing artificial intelligence models that can run on its own servers, free from remote provider control, a senior official announced Tuesday. The move reflects wartime Kyiv's drive to shield government services, businesses, and military tools from the risk of external shutdowns or restrictions.
Roman Kyslyi, Chief AI Officer at Ukraine's Ministry of Digital Transformation, told Reuters the policy favors self-hosted, or "on-premise," deployments. This approach limits reliance on models like those from Anthropic and OpenAI, which remain under vendor control.
The strategy gained urgency after the U.S. government ordered Anthropic to cut access to powerful models, a move that resonated across Europe. Kyslyi said the incident proved that AI sovereignty is not just rhetoric but a critical need.
"It confirms that AI sovereignty isn't just a defensive talking point, it's a necessity," Kyslyi stated.
Ukraine's stance comes as global tensions over AI access intensify. Reuters reported Tuesday that Chinese authorities are considering curbs on top open-source models, which dominate that market. For Kyiv, the decisive factor is not the model's origin but its deployability under Ukrainian control.
"If the vendor will provide it to run on our on-premise (infrastructure), there are no restrictions," Kyslyi explained. He added that the model itself is essentially a commodity, and Ukraine will work with any provider meeting its sovereignty requirements.
From Remote Dependency to On-Premise Control
Currently, Ukraine's AI assistant inside the Diia government app runs on Google's Gemini, accessed remotely through servers in the European Union. Kyslyi noted that Google provides free tokens, eliminating budget costs. However, Ukraine strips personal data before sending queries because it does not control those models.
Describing Gemini as an "interim" solution, Kyslyi revealed that Ukraine is developing its own model with telecom operator Kyivstar, based on Google's open-source Gemma. Set for release in autumn, it will serve government services, private enterprises, and the military.
The ministry evaluated several open-source options, including Mistral models and OpenAI's GPT-OSS, before choosing Gemma. Kyslyi said Gemma and Mistral matched remote-only alternatives on many performance tests, validating the shift to self-hosted AI.
This pivot mirrors a broader European push for digital sovereignty, accelerated by geopolitical shocks. Analysts note that the war has made Ukraine a real-world laboratory for resilient tech infrastructure. The approach also aligns with NATO's increasing focus on AI in defense, where reliance on foreign-controlled models poses strategic risks.
However, challenges remain. On-premise models require significant computational resources and expertise, which may strain Ukraine's wartime economy. Critics argue that open-source models, while flexible, may lag behind proprietary systems in cutting-edge capabilities. Yet for Kyiv, the trade-off is clear: control trumps convenience.
The Geopolitics of AI Access
The Ukrainian policy underscores a growing fragmentation in global AI governance. The U.S. move to restrict Anthropic's exports, coupled with potential Chinese curbs, signals an era where AI models become instruments of state power. Ukraine's stance could influence other nations facing similar dependencies.
Kyslyi's comments also highlight the evolving role of open-source AI. While often praised for transparency, models like Gemma are not immune to geopolitical pressures. Google's decision to keep Gemma open contrasts with its remote-only Gemini, offering Ukraine a path to independence.
Looking ahead, Ukraine's homegrown model could serve as a blueprint for other conflict-zone or privacy-focused governments. The autumn release will test whether on-premise AI can deliver at scale under extreme conditions. As Kyslyi put it, the model is a commodity—but control over it is not.
Reporting by Leo Marchandon and Gianluca Lo Nostro in Gdansk, Editing by William Maclean
