State Duma invents "sovereign" and "national" AI to hide stolen tech

The deputy chairman of the State Duma committee on information policy, Anton Tkachev, went on RBC Radio to officially split the country's AI future into two absurd buckets.

According to the new bill, a "sovereign" AI model must be completely developed within domestic borders using only local components. Because this is basically impossible in the current microchip climate, lawmakers graciously reserved this category exclusively for the military, police, and highly classified state agencies where "security is maximum" and performance is secondary.

For everyone else, they invented the "national" AI model, which is allowed to borrow foreign open-source code like LLaMA or Mistral. The only catch is that the servers must live physically in the country, and all local user data must be processed locally.

This desperate taxonomy is part of a sweeping bill targeting foundational AI models with more than one billion parameters. The parliament is still trying to define what a "parameter" actually is, since the term currently does not exist in local laws.

Both classes of models are promised priority access to government subsidies and critical infrastructure contracts starting September 1, with mandatory migration scheduled to kick off in March 2027.

It is a beautiful system: label the stolen Western tech as "national" to keep the economy running, call the non-existent fully domestic chips "sovereign" for the generals, and pretend everything is going according to plan. The transition period for the most critical sectors stretches all the way to 2032, giving everyone a decade to figure out how to run neural networks on wooden abacuses.