Ukraine is developing an independent large language model using Google's open-weight Gemma framework, aiming to create an autonomous artificial intelligence system serving military and civilian operations across the nation.
The project represents a strategic effort to achieve technological independence and reduce reliance on proprietary systems controlled by foreign companies, particularly following Ukraine's Deputy Minister of Digitalisation Oleksandr Bornyakov's determination in September that the country should avoid paying millions of dollars annually to foreign AI companies for essential capabilities.
Ukraine deliberately rejected OpenAI's ChatGPT, Chinese AI models including DeepSeek and Qwen, and other proprietary systems, choosing instead to leverage Google's open-source Gemma framework alongside an infrastructure partnership.
The initiative will initially use Google's computing infrastructure for training before transitioning entirely to local infrastructure, ensuring Ukraine retains full control over AI systems accessed by 23 million citizens daily through Kyivstar and government platforms.
The Ukrainian large language model project stems from practical limitations in existing AI systems' ability to handle Ukraine's linguistic diversity and specialized requirements.
Deputy Minister Bornyakov explained that when he communicates with people from his hometown of Bolhrad in Odesa Oblast, they use a blend of Ukrainian, Russian, and Bulgarian, a dialect that existing proprietary AI models struggle to process effectively.
The project addresses fundamental gaps, including mistranslations in legal documents, AI-generated hallucinations, and the inability to understand regional language variations used across Ukraine.
Ukraine's Minister for Digitalisation, Mykhailo Fedorov, noted in August that the military already employed AI for aerial and satellite reconnaissance and drone piloting, with tools from the U.S. firm Palantir Technologies helping analyze Russian strikes and monitor disinformation campaigns.
Why Ukraine rejected OpenAI and Chinese AI models for independence
Ukraine's deliberate rejection of proprietary AI systems reflects strategic calculations about technological sovereignty and military security.
Reliance on OpenAI's ChatGPT would create dependency on American corporate infrastructure and expose military operations to potential access restrictions or policy changes controlled by entities outside Ukraine's governance.
Chinese AI models presented additional concerns, with Deputy Minister Bornyakov explicitly stating they were rejected alongside OpenAI alternatives.
The choice to build independent capability rather than contracting with foreign providers ensures Ukraine maintains strategic autonomy in technologies increasingly critical to military operations and civilian governance.
The decision also reflects fiscal considerations, with Ukraine seeking to avoid millions of dollars in annual spending on proprietary AI services from foreign corporations.
Building local capacity using open-source frameworks enables Ukraine to achieve AI capabilities while retaining revenue that would otherwise flow to foreign companies.
This fiscal calculation becomes increasingly important as Ukraine faces substantial reconstruction costs and limited government resources.
By leveraging Google's open Gemma framework rather than proprietary systems, Ukraine achieves both technological independence and cost efficiency unavailable through commercial licensing arrangements with foreign AI companies.
Did you know?
Ukraine's AI project was driven partly by communication gaps where Deputy Minister Bornyakov's hometown dialect blends Ukrainian, Russian, and Bulgarian, which existing AI models struggle to process, demonstrating real-world limitations of proprietary systems.
How Google's Gemma framework powers Ukrainian digital sovereignty
Google's open-source Gemma framework was selected following extensive evaluation, with the company choosing it after considering alternatives such as Meta's Llama and France's Mistral AI models.
Gemma's performance across multiple languages proved decisive, with the framework's ability to handle diverse linguistic contexts making it ideal for Ukraine's multilingual population.
Google reinforced the selection by offering initial computing infrastructure for training, demonstrating a commitment to supporting Ukraine's digital independence while advancing the company's position in open-source AI development.
The infrastructure partnership enables Ukraine to access high-quality training resources without capital expenditures that smaller nations typically cannot sustain independently.
The choice of Gemma reflects practical technology assessment rather than geopolitical allegiance, with Ukraine explicitly evaluating and comparing options from multiple countries.
Google's openness in providing both the framework and infrastructure access created conditions enabling Ukraine to pursue independence while accessing world-class AI capabilities.
The framework's multilingual strengths address Ukraine's specific requirements in ways that more tailored systems might not provide.
Kyivstar's Chief Product Officer, Misha Nestor, noted that real-world production applications quickly reveal limitations in existing systems, suggesting that Ukraine's own testing identified gaps that make independent development necessary to meet national requirements.
The military and civilian applications driving Ukraine's AI development
Ukraine's military increasingly relies on AI for operations, including aerial reconnaissance, satellite imagery analysis, drone piloting, and battlefield management systems for troop coordination and enemy monitoring.
The Ministry for Digitalisation has implemented Palantir Technologies tools, helping analyze Russian strikes, monitor disinformation campaigns, and determine demining priorities.
An independent Ukrainian AI system would enable deeper integration of artificial intelligence into military command and control systems without dependency on external providers who might restrict access during critical operations.
The military applications represent perhaps the most urgent driver of Ukraine's AI independence initiative.
Civilian applications extend across government institutions, including courts, educational systems, regional archives, and public service delivery.
Data collection is underway from over 90 government institutions, establishing training datasets that reflect Ukrainian legal systems, educational content, historical records, and administrative functions.
The Ukrainian LLM will serve as a foundation for new generations of AI-powered services across the public and private sectors, with initial deployments supporting government applications and Kyivstar's platforms before expanding to broader private-sector use.
The comprehensive approach to both military and civilian applications ensures the AI system addresses Ukraine's strategic priorities while improving service delivery across government and commercial operations.
ALSO READ | Amazon and Google introduce joint cloud interconnect platform
What regional language support reveals about AI limitations
Deputy Minister Bornyakov's experience highlights fundamental limitations in proprietary AI systems' ability to handle linguistic diversity beyond major languages.
The Bolhrad dialect, combining Ukrainian, Russian, and Bulgarian, represents a communication reality for millions of Ukrainians living in regions with complex linguistic histories.
Existing AI models trained primarily on large language datasets from dominant speakers fail to process such regional variations, creating genuine gaps in practical applications.
Ukraine's project addresses these limitations by establishing advisory committees with binding authority over technical, linguistic, cultural, and historical aspects of the new model, ensuring it handles Ukrainian, minority languages, including Crimean Tatar, and Russian, widely used across Ukraine.
The linguistic requirements reveal how proprietary AI systems designed for global markets may overlook regional needs important to individual nations.
Companies like OpenAI prioritize maximizing user bases across global markets, potentially underinvesting in language support for smaller populations or regional variations.
Ukraine's independent approach enables prioritizing linguistic accuracy for its population and historical minorities, creating AI capabilities better aligned with national requirements than systems designed for universal global optimization.
The emphasis on linguistic accuracy extends to legal translation, where mistranslations could generate serious consequences for judicial proceedings and official documents.
Can Ukraine defend its AI system from Russian cyberattacks?
Deputy Minister Bornyakov explicitly acknowledged security threats, stating that Ukraine understands its AI system will face attacks immediately after release.
The ministry is developing defensive measures against prompt injection attacks, where malicious instructions embedded in tasks provided to the AI could compromise system integrity.
Russian cyberattacks continue targeting Ukraine's critical infrastructure, with energy infrastructure experiencing particular pressure.
Kyivstar, which became the first Ukrainian company to list on a U.S. stock exchange in August 2025, has installed more than 3,500 backup generators to stabilize operations as Russian strikes on Ukraine's energy infrastructure intensify.
Defending the AI system against sophisticated cyberattacks requires continuous monitoring, rapid response capabilities, and a security architecture hardened against attack vectors.
Ukraine's strategy involves deploying the system on secure local data centers after training on Google-provided infrastructure, potentially reducing exposure to attacks targeting cloud-based systems.
However, any publicly accessible AI system faces inherent security challenges from determined adversaries with significant resources.
Ukraine's preparation for inevitable attacks suggests a realistic threat assessment and commitment to maintaining system resilience despite anticipated compromise attempts.
The security architecture must balance accessibility, supporting military and civilian users with hardening against attack vectors while potentially facing sophisticated adversaries with a direct interest in disrupting Ukrainian government operations and military capabilities.
Comments (0)
Please sign in to leave a comment