Russia drafts its first major AI law: what changes for services, models, and platforms

Russia has introduced a draft of its first major framework law on artificial intelligence. The document does not directly ban chatbots or foreign models right now, but sets rules: when a citizen can demand a human instead of an algorithm, which systems will be allowed in state infrastructure, and who is liable for harm from AI.

Right to Refuse

The draft federal law "On the Basis of State Regulation of the Spheres of Application of Artificial Intelligence Technologies in the Russian Federation" was submitted for public discussion on March 18, 2026, and the discussion ended on April 15. If adopted, the document should enter into force on September 1, 2027.

The most notable rule for ordinary users is the ability to receive a service without autonomous AI if a person refuses such a format. However, this is not a universal right to "always switch to an operator": specific cases will be determined later by the government.

The document simultaneously introduces an obligation to warn if a product or service is sold with AI application without human involvement. A citizen must also be notified when an autonomous decision affects their rights, obligations, or legitimate interests. Even more important are two procedural guarantees: pre-trial appeal of decisions by state bodies, regional authorities, and companies with state participation that used AI, as well as the right to claim compensation for harm from unlawful use of such systems.

Which Models Will Be Allowed

The draft divides models into sovereign, national, and trusted. Sovereign and national models are strictly tied to Russian origin: they must be developed and trained in Russia, by Russian citizens and legal entities, on data generated in the country. This category is provided with state support and a separate regulatory regime. This is an attempt not just to describe the market, but to advance identify a segment that the state considers strategic.

A separate class is trusted models that can operate in state information systems and at significant facilities of critical information infrastructure. A registry, security requirements, data processing only on Russian territory, and mandatory quality inspection under government rules are proposed for them.

In parallel, the concept of cross-border AI technologies is introduced: such solutions may be restricted or prohibited in certain cases. This creates a legal mechanism for future decisions regarding foreign services, although the draft does not directly ban ChatGPT, Gemini, or Claude.

Marking and Responsibility

Another major section concerns synthetic content. Owners of AI services will have to explicitly mark audio, video, and other materials created with AI, and in a way that both humans and machines understand the marking.

For large platforms with more than 100 thousand users per day in Russia, requirements are stricter: if marking is missing, the service will have to add it itself or delete the material.

• Users must be warned about interaction with AI and about autonomous decisions affecting their rights
• Platforms must monitor the marking of synthetic content
• Developers, operators, and service owners are liable for unlawful results if they knew or should have known about the risk
• A user is liable if the violation resulted from their intentional actions
• Protection may extend to original results created with AI

The most controversial appears to be the intellectual property section. The draft proposes protecting results created with AI under the rules of the Civil Code and directly allows that original creations may arise not only with human participation but also in an automated system. At the same time, data extraction from objects protected by copyright or patent law for AI training is not considered a violation if a lawful copy is used or the material has already been lawfully published and is available for analysis. For rights holders and AI companies, this will likely become one of the main points of contention.

What It Means

Russia is transitioning from general talks about AI to assembling a complete legal framework. If the document is adopted in a form close to the current one, the market will receive not only new restrictions but also more understandable rules of the game: where a human is needed, which models are considered trusted, how to mark content, and against whom liability falls in case of an algorithm error.