Ukraine's Ministry of Digital Transformation has put a free, ten-episode training series called "AI for Public Service" on the Diia.Osvita e-learning platform, teaching civil servants how to use ChatGPT, Gemini, Claude, Perplexity, and NotebookLM for document analysis, prompt-writing, and routine correspondence. The course was built with the UN Development Programme under its DIA Support project, financed by the Swedish government, with sister civil-service literacy tracks on the same platform separately backed by Denmark, Switzerland, and Japan. It is a small program — ten short videos, 0.1 ECTS credits, no mandate to complete it — but it is a precise expression of where Ukrainian digital policy is headed.
From Digitizing Forms to Delegating Judgment
The course lands inside a much larger pivot the ministry has been narrating publicly since spring. At a Kyiv summit in May 2026, officials framed a shift from what they call the "digital state" — Diia, the app that has moved government ID, tax filings, and benefits online for more than 23 million Ukrainians — toward an "agentic state," in which AI systems don't just digitize a bureaucratic process but execute it. Diia.AI, an assistant already handling simple requests like income-tax certificates, is the citizen-facing piece; a Ukrainian-language model called Siaivo, trained on domestic registries and legal texts, is meant to eventually be the backend for both citizens and civil servants (Digital State). The civil-servant AI course is the workforce half of that bet: before you hand a bureaucracy AI agents to delegate to, you have to teach the humans supervising them what the tools can and can't do.
That sequencing is the right call, and worth stating plainly before getting to the catch. Ukraine's parallel investments back it up: the country climbed from 54th to 40th of 195 countries in Oxford Insights' 2025 Government AI Readiness Index, a 14-spot jump in a year of active war — the Cabinet of Ministers points out that Ukraine's digitized-services score already beats the US and UK on that metric. For a state simultaneously fighting an invasion and running one of the world's more advanced e-government stacks, training the workforce to use commercial LLMs competently, rather than banning them and letting staff use them unsupervised anyway, is the proportionate policy. A civil servant who understands prompt structure, knows to check outputs for hallucinations, and can tell when a document should not go into an AI tool at all is a better-governed civil servant than one improvising with no guidance.
The Gap the Course Doesn't Close
Here is the steelman for caution, and it deserves to be taken seriously rather than waved off: teaching thousands of officials to route "document processing and public administration tasks" through consumer versions of ChatGPT, Gemini, and Claude means routing an unknown volume of government records — personnel files, procurement drafts, citizen correspondence, potentially sensitive administrative material — through commercial US infrastructure outside Ukrainian jurisdiction, with no announced sovereign or contractually walled-off government tier for these three tools. The Diia.Osvita curriculum itself flags this: alongside prompt-writing, it teaches "data anonymization and privacy configuration" and how to spot AI hallucinations, which means the ministry knows this is the live risk, not a hypothetical one.
The risk is not abstract. In the United States, CISA's own acting director uploaded contracting documents marked "for official use only" to the public ChatGPT interface in mid-2025 under a personally requested exception — precisely the kind of well-intentioned shortcut that training alone doesn't fully prevent, since it was a senior, technically literate official who did it (CSO Online). Ukraine's exposure is arguably higher: its state agencies are under sustained, active targeting. On August 18, 2026, Ukraine's Asset Recovery and Management Agency reported a cyberattack alongside suspected unauthorized access to an internal personnel database, which it is investigating as a possible coordinated effort tied to a contested asset seizure (The Record). A government training tens of thousands of employees to feed working documents into third-party AI platforms, in that threat environment, is widening its attack surface faster than it is building the sovereign infrastructure — Siaivo, the AI factory with NVIDIA and HP — that would let it do the same work inside a controlled perimeter.
The Right Instinct, Sequenced Backwards
None of this argues for slowing down AI adoption in Ukrainian government — the productivity case is real, and a wartime state with a shrunk civil-service labor pool has more reason than most to automate routine document work. But training should ship paired with a data-handling policy that is enforced at the platform level, not just taught in a video: a government-tier agreement with the AI vendors, a hard rule on what document categories are off-limits for public tools, and a fast timeline for Siaivo to take over the sensitive slice of the workload. Kyiv has correctly bet that the biggest deployment risk is workforce ignorance. It should not let that bet obscure the smaller, sharper risk sitting right behind it: infrastructure that isn't sovereign yet, doing sovereign work, in the middle of a war.