MOSCOW, RUSSIA / RankWire.AI / – The Federation Council approved a national framework bill for artificial intelligence on July 17, setting rules for large foundation models in Russia. The measure defines covered technology and assigns powers to state bodies. It also introduces standards for model ownership, domestic data storage, user disclosure and AI-generated content. The bill passed the State Duma on July 8 and still needs presidential approval and official publication before it becomes federal law.
The proposal defines a large foundation model as software that can perform many intellectual tasks at a human-comparable level. A covered system must contain at least 1 billion parameters. It may provide information, make decisions or forecast results from goals set by people. The framework also establishes principles involving technological sovereignty, human rights, personal choice, security and compliance with Russian law. Those principles apply across the development, deployment and use of qualifying AI systems.
The legislation creates sovereign and national model categories tied to Russian control. A sovereign model must come from a Russian legal entity and use data centers inside the country. Its developer must retain the ability to reproduce the full development cycle, including training and original parameters. A national model follows similar ownership and localization rules. However, it may incorporate foreign software components released under open licenses, provided Russian entities retain the required control and operating capacity.
Domestic AI models receive legal classifications
The government may offer support to developers that build, deploy or operate qualifying foundation models. That support can include access to state-held datasets for training. Authorities may also require exclusive use of sovereign or national models in government information systems and other sensitive settings. Separate legislation or presidential decrees may establish additional rules for defense, security, public order and property protection. The framework gives state bodies responsibility for applying those requirements within their legal mandates.
Large digital services face a separate requirement for AI-generated audio and visual material. Services with more than 500,000 daily users must provide a tool that lets users mark such content. The provision covers websites, applications and social platforms. It does not require every item to receive an automatic label from the platform. Developers and users may define the notice format through service agreements. The requirement centers on making a disclosure option available to people who create or distribute qualifying material.
Copyright and disclosure provisions set standards
AI service providers must tell users who owns rights to generated material. They must also explain access conditions and whether users may download or transfer the content. The bill separately addresses copyrighted works used for machine learning. It permits analysis for extraction, comparison, classification and pattern detection when developers obtained lawful access. Training may use protected works when no one bypassed technical restrictions that controlled access to the material. The rules link model training to existing copyright and access conditions.
Most provisions are scheduled to take effect on September 1, 2026, after presidential signature and official publication. Rules covering domestic model status, developer duties, content marking and intellectual property will start on March 1, 2027. Existing systems may continue operating until September 1, 2032, when they process and store data inside Russia. Until the required signing and publication steps occur, the measure remains an approved bill rather than an enacted federal law under Russia’s legislative process.