The Russian government is launching a large-scale reform of science and technology development management, under which domestic artificial intelligence models will be trained exclusively on data collected in a unified state system. Prime Minister Mikhail Mishustin has approved a list of instructions following the strategic session, with the first results expected as early as October 2026.
This is not just about technical integration, but a fundamental paradigm shift: all research support measures, including tax incentives, will be moved into a single digital framework. The Ministry of Education and Science, the Ministry of Finance, the Ministry of Economic Development, and the Federal Tax Service must ensure a phased transition to the new system by November 27. This means that applicants and relevant agencies will work within one digital environment instead of fragmented procedures—a step that has long been needed to improve transparency and efficiency.
Data becomes a strategic resource
In parallel, the system is being populated with industry data. Chief budget administrators funding medical research are required to submit information on applied research for 2026. Medicine was chosen first not by chance: the MosMedAI platform already unites more than two thousand clinics across 75 regions and has processed over 16 million radiology studies. This is a colossal body of information that could serve as the foundation for training national models.
By November 25, the relevant agencies—the Ministry of Education and Science, the Ministry of Economic Development, Rospatent, and the Ministry of Digital Development—must work out a mechanism for training "sovereign" and "national" AI models on data from this system. The key requirement is robust protection of intellectual property. The documents clearly distinguish between two types of models: sovereign ones, hosted in domestic data centers without foreign components and operating in a closed loop, and national ones, which allow the use of open licenses provided they are hosted on Russian infrastructure.
Legal framework and copyright
Special attention is paid to copyright protection when training neural networks. The new mechanism will differentiate access to data sets to ensure the process does not devalue patent rights. The dispute over training data has already flared up: rights holders insist on transparent licensing terms, especially for scientific developments whose commercial value directly depends on their disclosure regime.
Mishustin emphasized that the new science management model requires close cooperation between the state, business, and scientific organizations—primarily in priority areas of scientific and technological development.
My view: this step is a logical continuation of the course toward technological sovereignty, but success depends on how quickly issues of data monetization and developer rights protection can be resolved. Without a clear legal framework, even the most ambitious initiatives risk remaining on paper.