Russia Just Passed an 'AI Law' to Force Models to Be Locally Hand-Crafted

The new regulation distinguishes between National and Sovereign AI models, specifically targeting large fundamental models with over 1 billion parameters. In a world where OpenAI and Google operate on trillions of parameters, setting the threshold at one billion is like regulating 'high-speed vehicles' by restricting the speed of a tricycle. The government now demands that Sovereign models be fully developed, trained, and hosted within Russia, ensuring the entire lifecycle—from the initial code to the final weights—is transparent and controlled by local entities.

These models must pass a compliance check for Russian 'traditional spiritual and moral values.' Essentially, this is a forced lobotomy for any AI operating within the country, ensuring that if you ask a question about sensitive geopolitics, the machine is hard-coded to parrot the party line. The law also mandates that National models can use open-source components, but Sovereign ones must be built from the ground up, with full technical reproducibility required to prove no 'hidden' foreign influence exists.

The regulation creates four major headaches for developers. First, if an AI service uses a foreign API like OpenAI or Claude, it technically processes data outside of the Russian jurisdiction. Second, sending documents or user queries to external clouds violates the strict data-locality rules. Third, even if a model is open-source, hosting it on AWS, Google Cloud, or Azure fails the requirement for local data centers owned by local firms. Finally, the 'black box' nature of APIs makes it impossible for developers to prove the 'full cycle of development' that the law now demands.

While the law doesn't explicitly ban foreign models yet, it grants the government the power to force specific sectors—like finance or public infrastructure—to adopt only 'certified' local alternatives. The industry is essentially being funneled toward YandexGPT and GigaChat as the only 'safe' options. It is a bold strategy to ensure that the only AI mistakes the population encounters are made locally, without any pesky, objective, or competent foreign data interfering with the state-approved narrative.