Russia has reached a crossroads in the creation of sovereign artificial intelligence. Experts from Lomonosov Moscow State University explicitly state: it is necessary to decide whether the model will be based on open-source solutions or on a completely domestic mathematical foundation.
The stakes are hundreds of billions of rubles. According to experts interviewed by the agency, writing a large language model from scratch requires colossal resources and time. This is why the government and relevant departments are discussing a compromise option using open source.
The State Duma has already laid the legal groundwork by adopting a law on the development of large fundamental AI models. The document introduces two key concepts: a "sovereign" model is developed exclusively by a Russian legal entity and operates only on Russian infrastructure, while a "national" model may include open-source solutions.
The main body of norms will come into force on September 1, 2026. Requirements for models, their accounting, developer responsibilities, content labeling, and the use of intellectual property results will become effective on March 1, 2027. The law also grants developers access to state data for training neural networks. The choice between expensive independence and rapid deployment based on open source will determine the architecture of the Russian AI market for years to come.
Read more on the topic:
- State Duma gives start to "sovereign AI": bill passed in first reading
- Mobile Internet in Russia to be Accelerated with AI: What Will Change for Subscribers
- AI under the law: State Duma prepares phased regulation of neural networks