Ukraine is building a large language model (LLM) using Google’s open-weight Gemma framework, aiming to create an independent artificial intelligence system for both military and civilian applications. The initiative, announced on Monday by Ukraine’s digital ministry and mobile operator Kyivstar, seeks to reduce the country’s reliance on foreign technology providers.
Initial training will use Google’s computing infrastructure before transitioning to Ukrainian servers. This move is intended to give Ukraine full control over AI systems used by 23 million citizens every day. Kyivstar said the model would underpin a new generation of AI-powered services across both public and private sectors.
Deputy Minister for Digitalisation Oleksandr Bornyakov said Ukraine wanted to avoid paying millions to foreign firms and instead invest in local development. He also confirmed that the country’s military plans to integrate AI into battlefield systems for troop coordination and enemy tracking.
Strategic Partnerships and Technological Goals
Google’s Gemma framework was chosen for its multilingual capabilities and performance. Ukraine’s Minister for Digitalisation Mykhailo Fedorov said the model builds upon existing AI work already in use for aerial and satellite reconnaissance and drone operations. AI tools from US-based Palantir Technologies are also helping Ukraine analyse Russian attacks and prioritise demining operations.
According to Kyivstar, the partnership with Google strengthens technological and economic ties between Ukraine and the United States — links that have grown since Kyivstar’s Nasdaq listing in August 2025. Alternatives such as Meta’s Llama and France’s Mistral AI were considered before Gemma was selected.
Building a Culturally Inclusive AI
Developers say the Ukrainian model is being designed to understand and generate multiple languages used within the country, including Ukrainian, Russian, Crimean Tatar, and minority dialects. Four advisory committees have been established to oversee technical, linguistic, cultural, and legal standards, ensuring the system reflects Ukraine’s diverse identity.
Data is being collected from more than 90 government institutions, including court registries, regional archives, and records documenting Russian actions during the ongoing war. Training will take place on secure GPUs hosted abroad, with final deployment planned for Ukrainian data centres.
Addressing Security and Reliability Challenges
Officials expect the system to face cyber threats upon release. Bornyakov said teams are developing safeguards against attacks, including “prompt injection” techniques that could manipulate AI behaviour. Kyivstar has already reinforced its infrastructure by installing over 3,500 backup generators to maintain stable operations amid continuing Russian strikes.
The first applications of the new AI system will serve government services and Kyivstar platforms, with private sector adoption to follow. Ukraine’s initiative highlights how smaller nations can use open technologies to strengthen digital sovereignty and reduce dependence on major global AI providers.
(with inputs from Reuters)
