Ukraine will give preference to artificial intelligence systems it can run on its own servers rather than solutions that depend on remote control by providers, a senior official at the Ministry of Digital Transformation said on Tuesday. The policy aims to keep digital tools used across government services, private firms and the military insulated from the risk that external providers could restrict or deactivate access.
The ministry’s approach emphasizes self-hosted - or on-premise - models that Kyiv can install within its own infrastructure. It explicitly steers away from solutions that remain under the provider’s operational control, a category that the ministry said includes some offerings from Anthropic and OpenAI.
Roman Kyslyi, Chief AI Officer at Ukraine’s Ministry of Digital Transformation, said the policy was reinforced by a recent U.S. government order for Anthropic to cut access to powerful models, and that this action echoed wider European concerns. "It confirms that AI sovereignty isn’t just a defensive talking point, it’s a necessity," he told Reuters.
The ministry does not base its decision primarily on a model’s country of origin. Kyslyi said the decisive criterion is whether a vendor will allow the software to run on Ukraine’s on-premise infrastructure. "If the vendor will provide it to run on our on-premise (infrastructure), there are no restrictions," he said.
Kyslyi described his view of the technology in plain terms: "The model is essentially a commodity," he said, adding that Ukraine would engage with any provider whose technology can be deployed under Ukrainian control.
At present, Ukraine’s AI assistant embedded in the Diia government application operates on Google’s Gemini model, which is accessed remotely through servers located in the European Union. Kyslyi said Google supplied free tokens for the Diia assistant, so the deployment incurred no budgetary cost. He also said the government strips personal data before sending queries to Gemini because those models are not under Ukraine’s control, and he described Gemini as an "interim" solution.
Separately, Ukraine is working with Kyivstar to develop its own model based on Gemma, the open variant tied to Google’s technology, which is scheduled for release in the autumn. That model is intended for use across government services, private companies and the armed forces.
When selecting a partner for its internally deployable model, the ministry compared several open-source options. Kyslyi said the review included Mistral models and OpenAI’s GPT-OSS, and that Gemma and Mistral performed comparably to remote-only alternatives on many evaluation tests.
The discussion of AI model access occurs alongside reports that Chinese authorities are also considering restrictions on top AI models, which currently have a prominent role in the open-source market, the ministry official noted.
Context and implications
The ministry’s policy places emphasis on technical control and deployability rather than on provenance, while the recent actions by other governments have underscored the operational risks of relying on provider-controlled models. Kyiv’s path combines an interim reliance on a remotely hosted Gemini instance with a parallel effort to field a Ukrainian-deployable alternative.