On 8 July 2026, the State Duma adopted, in its second and third readings, a law supporting the development of artificial intelligence technologies in Russia. The legislation has been submitted to the Federation Council.
The law introduces statutory definitions of artificial intelligence and a large foundation artificial intelligence model, establishes requirements applicable to sovereign and national AI models, and regulates state support measures available to their developers.
Scope of the Law
The law applies only to large foundation models: software containing at least one billion parameters, capable of performing a broad range of intellectual tasks and serving as a basis for the development or enhancement of other software products.
The regulatory framework is based on the principles of technological independence, protection of the rights and legitimate interests of individuals, security, support for Russian developers and a risk-based approach. The Government of the Russian Federation will be authorised to establish specific requirements for the use of models, taking into account the risks associated with their application.
Sovereign and National Models
The law introduces two special categories of large foundation models: sovereign models and national models.
A model in either category must be developed by a Russian legal entity. For the purposes of the law, this includes organisations controlled by the state, a municipal authority, a Russian citizen who does not hold foreign citizenship, or persons controlled by any of the foregoing. As a general rule, Russian companies under foreign control will not be eligible for the relevant status.
A sovereign model must be developed, and its entire technological cycle must be fully reproducible, by a Russian developer using its own resources. A national model may incorporate foreign components and other models, provided that they are distributed under an open licence.
For both categories, user requests must be processed, and data must be stored, in data centres located in Russia and owned by Russian legal entities. The model must also undergo a conformity assessment to confirm compliance with Russian legislation and traditional Russian spiritual and moral values.
The procedure for registering models, granting the relevant status and confirming compliance will be established by the Government of the Russian Federation. Until the necessary implementing regulations are adopted, the procedure for obtaining such status and the additional eligibility criteria remain undefined.
Support Measures for Developers
Developers of sovereign and national models will be eligible for state support. The law provides for financial, property-based, guarantee and information support.
It also establishes a legal basis for granting developers access to data contained in state information systems where such data is required to train models. The Government of the Russian Federation will determine the procedure for access to federal data. Access to regional data will be regulated by the authorities of the constituent entities of the Russian Federation in consultation with the competent security authority.
Developers will also be able to participate in specialised industry organisations, the development of ethical rules governing the use of AI, international research projects and the promotion of Russian technologies abroad.
The law separately provides for the possibility of insuring risks associated with the development and use of models, including liability for damage caused.
The practical significance of these provisions will depend on implementing regulations establishing the specific forms of support, the applicable eligibility conditions, the procedure for registering models and the process for granting sovereign or national status.
New Obligations for Developers and Digital Platforms
Developers of sovereign and national models will be required to implement organisational and technical security measures, establish rules governing the operation, updating and decommissioning of models, and maintain technical documentation describing their key parameters and limitations.
The law does not introduce a general obligation to label all materials created using artificial intelligence. Users must, however, be provided with the ability to add an appropriate notice. Major online platforms with a daily audience exceeding 500,000 Russian users must provide the technical functionality required to do so.
Persons providing access to models will be required to inform users who holds the rights to the generated outputs, as well as the conditions governing access to, use and retention of those outputs.
Use of Copyright-Protected Works for Model Training
One of the most significant provisions of the law is a specific exception to copyright and related rights for the training of sovereign and national models.
Computational analysis of patterns, trends and relationships contained in protected works, as well as the temporary recording of such works in computer memory for the purpose of model training, will not constitute an infringement.
The exception applies where the developer has lawfully obtained a copy of the work or where the work has been made publicly available and access to it is not restricted by technical means.
This provision broadens the ability of Russian developers to create training datasets. At the same time, technical measures for managing access to materials published online will become increasingly important for rights holders.
Implications for Companies Using AI
The Government of the Russian Federation will be authorised to determine the circumstances in which only sovereign or national models may be used. In the banking and financial sectors, such decisions must be agreed with the Bank of Russia.
The relevant industries and specific use cases have not yet been identified. The consequences for companies using foreign models or solutions based on them will therefore depend on future regulations adopted by the Government of the Russian Federation.
A transitional period until 1 September 2032 is provided for information systems created or operated before the relevant requirements enter into force, provided that the data is processed and stored in Russia.
Entry into Force
The principal provisions of the law are expected to enter into force on 1 September 2026.
The following provisions are expected to take effect from 1 March 2027:
- circumstances in which the use of sovereign or national models will be mandatory;
- requirements applicable to such models and the procedure for granting the relevant status;
- obligations of developers;
- labelling of materials;
- use of protected works for model training;
- access to data contained in state information systems.
Before most of the new requirements can be applied in practice, the Government of the Russian Federation must establish the procedure for granting model status, the conditions for state support, the rules governing access to data and the requirements for preventing and mitigating risks.
The law’s principal impact on the market will therefore depend on the content of the implementing regulations adopted pursuant to it.
