Law on Artificial Intelligence: Basic rules for regulating large fundamental models approved

The President signed a law that comprehensively regulates the development, implementation, and use of large-scale fundamental models of artificial intelligence (LFMs) for the first time. Most of the provisions will come into force on March 1, 2027. The document defines which models will be considered national and sovereign, establishes requirements for their developers, provides for the possibility of limiting the use of certain AI solutions in certain areas, and introduces rules for labeling content created using artificial intelligence. In essence, the law forms the basic legal basis for the further development of the domestic AI ecosystem and enshrines the state’s approach to regulating the most powerful generative models.

A large fundamental model is a software system capable of performing a wide range of intellectual tasks, including:

• information analysis,
• text generation,
• images and program code,
• forecasting and decision making based on machine learning.

Such models are used not only as standalone products but also as a technological foundation for creating various digital services and specialized application solutions. One of the mandatory characteristics of a BFM is the presence of at least one billion parameters, which effectively places the most modern generative artificial intelligence models in this category.

Requirements for Russian models

The law distinguishes two categories of domestic models: sovereign and national. Both categories are subject to uniform basic requirements.

In particular:

• the developer must be a Russian legal entity;
• computing infrastructure and data processing centers must be located on the territory of the Russian Federation and belong to Russian organizations;
• The model must undergo a procedure to confirm its compliance with the requirements of Russian legislation and the criteria established by the state.

However, differences remain between the two categories. Under the sovereign model, the Russian developer must fully control the entire product development lifecycle and be able to independently replicate the development process. Under the national model, the use of components distributed under open licenses, including those of foreign origin, is permitted; however, the key characteristics of such a model must be defined and modified by the Russian developer.

Use of models in specific areas

One of the most significant amendments to the law is the granting of authority to the government to determine cases in which the use of exclusively sovereign or national large-scale fundamental models is permitted. It is anticipated that such restrictions will be imposed primarily in areas of strategic importance, where issues of information security, data protection, and technological independence are particularly crucial.

The law also provides for a transition period. The restrictions will not apply to information systems created or commissioned before March 2027, provided that data processing and storage are carried out within the Russian Federation. This approach will allow organizations to gradually adapt to the new requirements without the need to immediately replace existing digital solutions.

Labeling AI-generated content

A separate regulatory framework is dedicated to the transparency of the use of artificial intelligence technologies. The law provides for the labeling of visual and audio content created using large-scale fundamental models. Specific requirements for the form and method of user notification will be determined by the terms of use of the relevant services.

Furthermore, social media owners must provide users with the technical ability to independently label materials created using artificial intelligence. It is expected that such labeling will increase the transparency of the origin of digital content and reduce the risk of disseminating materials that could mislead users.

Practical implications for business

Although most of the law’s provisions will only come into effect on March 1, 2027, organizations using artificial intelligence technologies should consider the new requirements now when selecting digital solutions and planning IT projects. The changes will primarily affect AI platform developers, software providers, digital service owners, and companies actively using generative artificial intelligence in their operations.

Businesses are advised to evaluate their AI tools in advance, as Russian models may be prioritized in certain sectors. Furthermore, organizations creating text, graphic, or audio content using generative AI should consider the prospect of mandatory labeling mechanisms.