Ukraine’s government has moved from digital services to AI-driven ones, and more is planned. HROMADA member Dmytro Khutkyy explains what has been launched, what is still promised and where the challenges lie.
Introduction
Ukraine’s Ministry of Digital Transformation no longer describes its goal as “digital-by-default.” It now speaks of an “AI-native” or “agentic” state. The Ministry of Digital Transformation sees the agentic state as resting on three pillars: services, human capital and adaptive regulation. In April 2026, acting minister Oleksandr Bornyakov described an experimental system that analyses all government decisions and labels their effects, while admitting its outputs cannot yet be fully relied upon.
The flagship is Diia.AI, an assistant in Diia, Ukraine’s e-government portal for digital documents and public services. It obtains documents and records on a citizen’s behalf rather than merely answering questions. Behind it stand a national language model, an action plan for artificial intelligence and training programmes for civil servants. Why does this matter beyond Ukraine? Because workers and public money are scarce at once, automation is a necessity, and Ukraine is testing at scale what other states may face in the future.
From mobile app to AI assistant
Diia.AI was launched in September 2025. On 4 November 2025, at the WINWIN Summit in Kyiv, the Book of World Records recognised it as the first AI assistant providing public services at a national level.
The design choices deserve attention. According to Ukraine’s Ministry of Digital Transformation, the assistant relies on two components. The first is Liquio, a low-code platform that standardises the logic of public services. The second is the Model Context Protocol, which describes to the AI which services exist, which parameters they need and what result to expect.
The Ministry also states that “AI does not see personal data”: identifiers are filtered out before a request reaches the language model, and the Diia system adds the personalised answer locally after the user has been authorised. This principle is called privacy-by-design. It should be noted that the implementation of this principle in practice rests not on an independent audit, but on the Ministry’s own description.
The Diia.AI assistant is also changing in form. On 21 July 2026, voice access was introduced, which allows users to speak, hear a reply and switch between voice and text within one conversation. Reportedly, the wider Diia ecosystem now embraces more than 24 million users and over 170 digital public services.
Sectoral AI beyond Diia
Not every AI tool sits inside Diia. The State Tax Service announced an AI agent for its taxpayer account, due by the end of 2026. The Ministry of Justice pilots a legal chatbot that answers from official sources, and administrative service centres test an assistant for staff.
The stakes and automation levels differ. Victoria Shi, a digital spokesperson, handles routine consular information. Cassandra, in use since 2020, estimates reoffending risk for pre-trial and pre-sentence decisions. The National Agency on Corruption Prevention’s checks of declarations are rule-based automation, not AI.
The language model behind the state
An AI assistant needs a model that understands legal Ukrainian. The principal solution here is the national language model. In late 2025, Kyivstar, Ukraine’s largest mobile operator, and the Ministry of Digital Transformation chose Google’s open Gemma family as the base. Kyivstar finances and leads the technical work, and according to the Ministry, no public budget funds are used at this stage. Training texts come from government bodies, media, universities, scientific institutions, and historical archives. The Ministry itself admits that “a significant share of relevant Ukrainian-language material still exists only in offline or paper form.”
The model was named by public vote. On 30 March 2026, “Syayvo” won with 22,600 of more than 136,000 votes and became the official name of Ukraine’s national model. An open-source release is expected after validation. Whether the model will manage regional dialects and the languages of minorities remains an open question, and an important one for legal nuance and for inclusion.
Additionally, there is a local language model. In September 2025, the INSAIT institute, working with the Ukrainian Ministry of Digital Transformation and the Bulgarian Ministry of Innovation and Growth, released MamayLM v1.0. Its developers position the model for education, law, healthcare and the public sector, with local deployment to protect data.
Language technology extends to handwriting. In May 2026, RUKOPYS, Ukraine’s first large open annotated dataset of Ukrainian handwriting, was announced to help services process handwritten applications, registers and certificates.
Ukraine thus pursues two logics at once. A centralised solution, with one national model (Syayvo) and one framework (Diia), brings consistency, shared standards and convenience for citizens and officials. A set of models and independently launched applications, from MamayLM to ministry-level pilots, diversifies the field, spreads risk and lets startups and agencies experiment, but it fragments data, quality control and accountability.
Opening the ecosystem to startups
Since June 2025, the Ministry of Digital Transformation, the Ukrainian Startup Fund, and the UK-funded UK DIGIT project have run an innovation sandbox for companies developing AI or blockchain products in public services, health, agriculture, education, defence, and infrastructure. The programme runs through October 2026, with plans to extend and formalise it if it proves successful.
The sandbox complements the government’s Action Plan for the Concept of AI Development for 2025–2026. A regulatory choice underpins it: Ukraine tests first and legislates with evidence from the sandbox, in a setting that aims to align with the European Union.
Diia.Business launched five AI assistants in December 2025 that guide entrepreneurs from idea to funding and export, but, the Ministry of Digital Transformation states, do not replace expert consultations.
Automation is not optional, but essential
The labour market explains the urgency. According to the Centre for Eastern Studies (OSW), 67% of Ukrainian enterprises reported labour shortages in March 2026, more than the number who named security or power outages as their main problem. Official employment fell to 10.7 million by the end of 2025, 1.3 million below the pre-war level. As a result, public administration competes for the same scarce people.
AI offers the state a way to deliver more with fewer bureaucrats. UNDP reports, for instance, that the Ministry’s lawyers cut document verification from weeks to 72 hours, its HR team automated 45% of processes, and the Diia.AI assistant now independently resolves over 95% of user queries. Such results are encouraging, yet they need external verification.
Reconstruction: when DREAM meets AI
An earlier post on the DREAM e-system recommended that international donors channel reconstruction funding through it. Since then, the system has changed in scale and in status. According to the Open Contracting Partnership in June 2026, its pipelines contained more than 14,000 projects with an estimated value of UAH 1.26 trillion, across more than 900 communities and 23 regional administrations.
Moreover, use is no longer optional. Parliament amended the Budget Code in January 2025, and funding decisions now require projects to be present in DREAM. The platform also uses a retrieval-augmented generation (RAG) assistant based on the local MamayLM model in an isolated environment, which supports chat help, contextual prompts, and checks of feasibility studies.
Yet the Open Contracting Partnership is frank about the limits. Small communities struggle with procedural requirements, and numerous first-cycle submissions amounted to “an exercise in submitting data to meet system requirements rather than doing the analytical work the reform intends.” Software cannot replace analytical capacity in a small community.
The Register of Damage for Ukraine, with claims filed through Diia, allows AI in mass-claims processing if the Register’s Board approves, and its Secretariat rechecks every such claim.
Digital skills of general users and civil servants
Technology moves faster than skills. On 3 August 2026, the Ministry of Digital Transformation and UNDP, with Swedish funding, launched a ten-episode series on the Diia.Education platform called “AI for Public Service”. Over-reliance is a known risk of public-sector AI, and a bureaucrat who trusts a fluent but wrong answer can do more harm than a slow bureaucrat.
Sectoral tools also change daily work. The Mriia platform pilots an AI test generator with a double AI check, though no human review is described. The State Employment Service’s voice assistant answered nearly 2,700 calls, roughly one every 18 seconds, during its pilot run, including during air raid alerts.
Digital literacy among ordinary users is the other half. A UNDP-supported survey by the Kyiv International Institute of Sociology found that 59% of Ukrainians used government e-services in autumn 2025, up from 55% a year earlier, and the share of non-users citing a lack of skills fell from 44% in 2024 to 22%.
Headline figures can hide gaps, however. The Ministry’s own study reports that 96% of Ukrainians have digital skills, yet only 58% reach basic or higher proficiency, and only 3% of visually impaired respondents report above-basic skills. A voice assistant helps those who struggle with forms, but only if they trust the service, know what they have consented to and can reach a human when the assistant fails.
Rollout prospects and risks
AI deployment risks are not hypothetical. Models trained on public-sector texts can inherit bias from past decisions. Over-reliance can erode human judgement in the very offices that must oversee exceptions. Adversarial attacks, including prompt manipulation, become more serious when an assistant can retrieve official records and trigger services. Ukraine’s cyber defenders have learnt from a decade of Russian attacks, as argued in the earlier blog post on cyber resilience, but an AI layer widens the attack surface. Finally, accountability grows complicated when a commercial model, a telecom-financed national model and a state platform all take part in one service. Table 1 summarises progress.
| Component | Announced objective | Status as of October 2026 |
| Diia.AI assistant | Move from information to executing services | Live since September 2025; voice added July 2026 |
| National language model | Ukrainian model for the public sector | Named Syayvo, open-source release pending |
| Sectoral AI | Ministries run their own tools | Tax agent planned; justice and service-centre assistants in testing |
| Open ecosystem | Startups build on state infrastructure | Sandbox runs until October 2026; extension pending |
| Reconstruction (DREAM) | AI support for project preparation | RAG assistant in use; DREAM presence required for funding |
| Civil service skills | Training for officials | Ten-episode Diia.Education series since August 2026 |
Table 1. Announced goals and statuses of Ukraine’s AI initiatives
Conclusions and recommendations
Ukraine uses its AI tools to deliver better public services. Donors who fund parts of this effort, such as the sandbox and the training series, are in effect co-authors of Ukraine’s digital state, giving them a stake in good governance. Yet Ukraine is not in the EU, so the EU AI Act and other legislation do not bind its domestic use of AI; it relies instead on a bottom-up, adaptive approach. This approach leaves more room for innovation but makes the ecosystem less consolidated and less subject to ethical checks.
Four steps would make the government’s achievements credible beyond its own announcements. First, the government should state national model language coverage openly, including what the model will and will not propose for the Crimean Tatar language and the dialects of the Ukrainian language itself. Second, the Ministry of Digital Transformation should commission independent evaluations of Diia.AI and Syayvo, including error rates and user surveys run by bodies outside government. Third, the sandbox should be extended and linked to clear rules on procurement, cybersecurity, and ethics, so that startups know what they are building towards. Fourth, Ukraine should gradually align its AI regulation with the EU framework, beginning with high-stakes uses, so that room for innovation does not come at the cost of ethical checks.
An agentic state is a promise until it is measured. Ukraine has delivered more than most governments, and it has done so under war. The next test is whether it can show, with open data and outside scrutiny, that the AI initiatives make the delivery of public services faster, fairer, and safer, and that they are transforming Ukrainian public administration itself.
October 2026. Text (excluding pictures): Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International license.
