Ukraine has begun developing a sovereign “AI brain” for both military and civilian use, aiming to reduce reliance on foreign models and retain control over sensitive data.
The project, led by the government in partnership with Ukrainian firm Kyivstar, will use Google’s open-source Gemma 3 large language model as the foundation.
According to Minister of Digital Transformation Mykhailo Fedorov, the initiative brings Ukraine closer to becoming a leader in AI by 2030.
“The quality and safety of our public services will directly depend on the quality of this model,” he added.
Initial training will take place on Google’s cloud infrastructure abroad before the system is deployed on Ukrainian servers.
The rollout is expected to face immediate threats, such as prompt injection attacks and potential disruptions from Russian strikes on energy infrastructure.
In response, protections for AI safety operational continuity are also in the works, including plans to install more than 3,500 backup generators.
Choosing Gemma
Ukraine evaluated other AI frameworks, including Meta’s Llama, France’s Mistral AI, and Chinese models like DeepSeek and Qwen, but all of them were rejected in favor of Gemma.
Fedorov cited its efficiency, multimodal text-and-image processing, and built-in Ukrainian language support.
The model will be further trained on Ukrainian-specific data, including minority languages like Crimean Tatar.
Sources from government institutions, court registries, educational publishers, regional archives, and records of Russian military actions will also reportedly be fed into the system.
Ukrainian Military AI Use
Ukraine’s armed forces are already using AI to support a wide range of missions, from aerial and satellite reconnaissance to drone operations and battlefield analysis.
Current tools include an AI-powered mothership drone, intelligence-gathering accelerators, and “Clarity” software used to spot Russian targets in seconds.
