Ukraine Advances in Artificial Intelligence Development
December 1 (Reuters) – Ukraine is embarking on a significant initiative to develop a sophisticated language model utilizing Google’s open-weight Gemma framework.
The endeavor aims to establish an autonomous system for artificial intelligence functions, responding to an escalating demand within both military and civilian domains, as reported by the nation’s digital ministry and mobile operator, Kyivstar.
The project will initially leverage Google’s computing infrastructure for training purposes, subsequently transitioning to a fully local infrastructure. This transition is pivotal, ensuring that Ukraine maintains complete control over AI systems utilized by approximately 23 million citizens on a daily basis.
“The upcoming Ukrainian LLM is anticipated to become the cornerstone for a new era of AI-driven services across public and private sectors in Ukraine,” Kyivstar articulated in a statement.
In a bid to curtail dependency on foreign entities, Ukraine seeks to cultivate its own AI model rather than incurring exorbitant costs associated with international firms. Deputy Minister for Digitalisation Oleksandr Bornyakov previously communicated to Reuters the rationale behind this strategic choice.
Bornyakov emphasized the intention to avoid reliance on proprietary systems, such as OpenAI’s ChatGPT, explaining that the military envisions integrating AI into battlefield management systems, facilitating troop coordination and enemy surveillance. Additionally, Chinese language models, including DeepSeek and Qwen, were also dismissed as options.
Ukraine’s Minister for Digitalisation, Mykhailo Fedorov, disclosed in August that the military has already implemented AI technologies for aerial reconnaissance, satellite monitoring, and drone operations.
Tools from U.S. company Palantir Technologies are currently aiding in the analysis of Russian offensives, monitoring disinformation campaigns, and prioritizing demining operations.
Gemma was selected for its robust performance across multiple languages, as indicated in the documentation shared by the ministry.
“Google was chosen following a thorough evaluation process, which further solidifies the technological and economic affiliations between Ukraine and the United States—an alliance that was enhanced by Kyivstar’s Nasdaq listing in August 2025,” stated Kyivstar on Monday.
Alternative AI models, such as Meta’s Llama and France’s Mistral AI, were also under consideration, according to two sources familiar with the developments.
This project addresses existing communication deficiencies in current AI systems. Bornyakov illustrated the challenge, noting that residents of his hometown, Bolhrad in Odesa Oblast, often communicate in a mix of Ukrainian, Russian, and Bulgarian—a blended dialect that prevailing AI models struggle to interpret accurately.
“The limitations become apparent when deploying real-world applications,” remarked Misha Nestor, Kyivstar’s chief product officer overseeing the initiative. He pointed out errors like misinterpretations in legal documents and AI-generated “hallucinations.”
The consortium has convened four advisory committees endowed with binding authority over technical, legal, cultural, historical, and linguistic dimensions of the emerging model.
This structure ensures the model adeptly manages Ukrainian and minority languages, including Crimean Tatar, alongside the widely used Russian language within Ukraine.
Data collection efforts are in progress, aggregating information from over 90 government institutions, encompassing court registries, educational publishers, regional archives, and documentation of Russian actions during the ongoing conflict.
Training will take place on secure graphics processing units located outside Ukraine, provided by Google, prior to deploying the finalized models on domestic data centres. Kyivstar has yet to confirm the anticipated launch date.
As the rollout approaches, security concerns loom large amid persistent Russian cyberattacks. “We recognize that an immediate onslaught is to be expected following the release,” asserted Bornyakov. Other AI services have similarly faced targeted assault, he added.
Preventative measures against potential threats, such as “prompt injection” attacks—malicious directives embedded discretely within tasks assigned to the AI—are currently under development.
Kyivstar, which made history as the first Ukrainian entity to list on a U.S. stock exchange in August, has installed over 3,500 backup generators to fortify operations amid intensifying Russian assaults on Ukraine’s energy infrastructure.
Initially, the completed AI system will support governmental applications and Kyivstar’s platforms, with plans for later expansion into the private sector.
Officials maintain that this project exemplifies how smaller nations can harness open technologies to achieve strategic autonomy in the AI arena, countering dependency on dominant foreign systems.
Source link: Thestar.com.my.
