Russia’s Federation Council has approved a comprehensive law governing the development, deployment, and oversight of artificial intelligence systems, setting the stage for a new legal framework around large-scale AI models. The legislation, passed on Wednesday, establishes two distinct categories for so-called big foundational models—those with at least one billion parameters and performance matching or exceeding human capabilities.

The law introduces a sharp distinction between “sovereign” and “national” foundational models. A sovereign model must be built by a Russian legal entity, with full control maintained across every stage of its lifecycle, from training to deployment. It must also be hosted exclusively in domestic data centers. A national model, by contrast, can incorporate foreign components under open licenses, though both types are required to comply with Russian law and traditional values.

Developers of these models will qualify for government support including financial, property, and information assistance. Crucially, they will be permitted to train neural networks using data from federal and regional state information systems. The specific rules for accessing that data will be set by the government in coordination with security agencies, a move that balances innovation incentives with state oversight.

Source: RIA Novosti