Ukraine’s public administration continues to operate under wartime conditions that place constant strain on public institutions. Staff shortages, budget pressure and damaged infrastructure limit the capacity of government bodies, while citizens still depend on the uninterrupted delivery of essential public services.
Administrative authorities must process large volumes of documentation, respond to citizen requests and maintain everyday procedures with limited human resources. In this context, purely manual processes increasingly struggle to keep pace, creating risks to service availability and response times.
Why language matters for agentic AI in public administration
To address these constraints, Ukraine is exploring AI systems that can support administrative work, assist citizens and help maintain service continuity when human capacity is stretched. In practice, this increasingly points toward agentic AI systems — AI that can support multi-step administrative processes, guide users through procedures and assist officials with routine workflows, rather than simply generating text.
For this to be viable in the public sector, language performance is critical. AI systems must operate accurately in Ukrainian, particularly when dealing with legal and administrative texts. Many global AI models are primarily trained on English and provide limited visibility into their data sources or filtering methods, making it difficult for public authorities to assess their suitability for sensitive or high-stakes use.
Evaluating European language-focused AI foundations
Against this background, Ukraine is evaluating European-developed AI models as alternatives to global, English-centric systems, particularly for language-critical public-sector use. These models are designed with public-sector requirements in mind, including language accuracy, transparent development practices and the ability to be deployed within national infrastructure.
One of the European AI models being assessed in this context is TildeOpen, a 30-billion-parameter open-source language model developed by Tilde and trained on Europe’s LUMI supercomputer. TildeOpen supports 34 languages, including Ukrainian, and has been developed to meet the needs of language-intensive and security-sensitive environments.
A first step toward trusted European AI
This direction is reinforced by a Memorandum of Cooperation signed by the Ministry of Digital Transformation of Ukraine, Latvia’s National Artificial Intelligence Centre, the Government Coordination Office for European and Euro-Atlantic Integration, and Tilde. The agreement marks a first concrete step toward developing trusted European AI foundations for public administration and citizens, with a focus on language accuracy and transparent model behaviour.
Tilde’s CEO, Artūrs Vasiļevskis, emphasises that Ukraine’s work carries wider significance for Europe: “Ukraine is strengthening its public services in circumstances where others would struggle to maintain even basic continuity. The decision to invest in reliable and linguistically precise AI reflects a clear understanding of modern public administration needs. Trustworthy AI are essential not only for Ukraine, but for Europe as a whole.”
Ukraine’s approach signals not only an immediate response to wartime constraints, but a long-term vision for the transformation of public administration. The government is developing plans to significantly extend digital public services, with the aim of making a broad range of administrative procedures accessible to citizens through agentic AI-driven services. In this context, the preference for AI developed in the European Union — grounded in European values of transparency, accountability and public trust — is a strategic choice. As Ukraine aligns its digital governance with European standards, language-focused and value-aligned European AI is positioned to play a central role in shaping the future of resilient, citizen-centred public services.