MOSCOW, RUSSIA / RankWire.AI / – Russia’s Federation Council has endorsed a legislative bill on July 17 that defines the national standards for the development and deployment of large artificial intelligence foundation models. The legislation introduces legal terminology, delegates regulatory authority, and provides support mechanisms for qualified domestic developers. It also outlines requirements related to model oversight, data storage, user notifications, and AI-produced content. Having been approved by the State Duma on July 8, the bill now awaits presidential approval before it becomes law.
According to the bill, a large foundation model refers to software capable of performing multiple intellectual tasks at a level comparable to human performance. This includes systems that generate information, make decisions, or predict outcomes based on human-set objectives. To qualify, a model must contain a minimum of 1 billion parameters. The framework also establishes guiding principles centered around technological sovereignty, human rights, personal choice, security, and adherence to Russian legislation.
The law creates distinct categories for sovereign and national models. A sovereign model must originate from a Russian legal entity and stay under domestic control, utilizing data centers within Russia. Developers are required to maintain the technical capability to reproduce the entire development cycle, including training and original parameters. A national model adheres to similar ownership and data processing rules but may incorporate foreign components licensed under open terms.
Domestic models receive formal recognition
The government can introduce support programs for firms involved in creating, deploying, or managing qualifying foundation models. Such initiatives may include access to government-held datasets essential for training purposes. The legislation also authorizes authorities to designate specific environments where only sovereign or national models are permitted, particularly in government information systems and other sensitive sectors. Additional regulations and presidential decrees might regulate areas related to defense, security, public order, and property protection.
Websites, applications, and social media platforms will face a new content regulation. Platforms with more than 500,000 daily users must implement a tool allowing users to flag AI-generated audio or visual content. This enables users to add informational labels to qualifying material. Developers and users will determine the format of these labels through service agreements. Platforms are not required to automatically label every piece of content but must provide the technical means to do so.
Copyright considerations and implementation timelines
AI service providers are mandated to disclose to users the ownership rights concerning generated content. They must also clarify access conditions and whether users can transfer the material. The legislation separately addresses copyrighted works used for machine learning purposes, considering analysis for tasks like extraction, comparison, classification, and pattern recognition lawful when developers have obtained legal access. Training on protected works is permitted when no technical restrictions were bypassed to access the material.
Most provisions are set to take effect on September 1, 2026, pending presidential approval and publication. Regulations concerning domestic model status, developer responsibilities, content labeling, and intellectual property will commence on March 1, 2027. Existing systems will be granted a transition period until September 1, 2032, during which their data processing and storage must remain within Russia. Until the measure receives presidential approval and official publication, it remains a proposed bill rather than an enacted law.