Russia's AI law favors state control over frontier capability

Russia's first law governing large foundation AI models took effect on September 1, 2026. President Putin signed it on July 26, after a broader spring draft was replaced in June by a narrower bill focused on large foundation models and state support. Writing in Tech Policy Press, Dmitry Kuteynikov and Lyubov Popovets argue the result is less a risk framework than an industrial-policy instrument.

The law creates two model statuses, "sovereign" and "national," requiring Russian legal entity ownership, data localization within Russia, and compliance with "traditional Russian spiritual and moral values." Models with at least 1 billion parameters count as large foundation models. Beginning in March 2027, a copyright exemption on training data will apply to sovereign and national models only. It is a status-tied carve-out unavailable to any other developer.

"The law does less to manage the risks of AI than to establish a framework for supporting and controlling developers of large foundation models," the authors write.

Four companies lead development inside Russia: Sber, T-Bank, Yandex, and MTS. All depend on state licenses or government approval, and their flagship systems lean heavily on foreign architecture. Sber's GigaChat is modeled on China's DeepSeek, YandexGPT 5 Pro used Qwen-2.5-32B weights, and the T-Bank and MTS systems build on successive Qwen generations. On rights, Kuteynikov and Popovets are blunt: the law provides "no prohibitions, no human rights impact assessment, no independent oversight, no notification of affected persons."

The design is already being marketed abroad. Sber has proposed a BRICS platform for AI sovereignty aimed at states in Africa, Asia, Latin America and Oceania "that want their own AI but cannot afford to build it." What is exported, the authors conclude, "is therefore not just AI capability, but a model of technological control." Two of the researchers in our Who's Who tracker shared the piece this week.