TL;DR
Russia's first comprehensive AI law, signed July 26 and in force September 1, 2026, trades state support and market protection for developer compliance with Russian ownership rules, data localization, and ideological conformity requirements.
What happened
- Russia's "On Support for the Development of AI Technologies" law took effect September 1, 2026, after President Putin signed it July 26.
- The law targets "large foundation models" defined as models with at least 1 billion parameters, covering both domestic developers and foreign providers serving Russian users.
- Two official statuses, "sovereign" and "national," unlock state support in exchange for Russian legal entity registration, Russian-owned data center storage, and conformity with "traditional Russian spiritual and moral values."
- A March 2026 draft was far broader, covering individual rights, governance, and IP; heavy criticism from lawyers, businesses, and government bodies forced a complete rewrite by June.
- The Ministry of Digital Development, FSB, and FSTEC are drafting the assessment procedure, under which developers disclose architecture and filtering mechanisms for accredited lab testing.
Why it matters
- The law institutionalizes state leverage over Russia's four dominant foundation model developers: Sber, Yandex, T-Bank, and MTS, all of which already depend on state licenses and contracts.
- Russia's AI stack is heavily foreign-dependent: YandexGPT 5 Pro was built on Qwen-2.5-32B weights; T-Bank's T-Pro and T-Lite and MTS AI's Cotype are built on successive Qwen generations, exposing a structural vulnerability the law does nothing to fix.
- Hardware dependence compounds the risk: Western export controls block advanced accelerators, forcing reliance on Chinese chips and pre-sanction or smuggled Western GPUs.
- The 1-billion-parameter threshold is a crude instrument compared to the EU AI Act's compute-based criteria (10^23 FLOPs for GPAI, 10^25 FLOPs for systemic risk), meaning the law captures small open models and frontier systems alike without meaningful risk differentiation.
- Requiring ideological conformity assessments with no defined procedure gives the FSB and FSTEC broad discretionary power over what models can legally operate in Russia.
What to watch next
- Government resolution on assessment procedures: the rules for granting "sovereign" and "national" status are not yet published; the FSB and FSTEC's final role will determine how coercive the regime becomes in practice.
- Foreign developer response: the law explicitly covers models made available to Russian users, so Western and Chinese providers must decide whether to comply, geo-block, or exit the market.
- Domestic hardware progress: Russia's own accelerator designs are described as years away; any acceleration in that timeline would reduce the structural dependency that currently undermines the law's sovereignty ambitions.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.