← Back

Russia's AI Bill Demands Spiritual Chatbots and Scraps Copyright Rules

Original version ·

Just when you thought AI regulation couldn't get more unhinged, northern lawmakers decided to drop legal safety nets and substitute them with state-mandated ethics. Because who needs actual compliance when your neural network loves the Motherland?

The completely rewritten draft law, originally published by the Ministry of Digital Development in March 2026, has chopped the bill from 21 articles down to 13. Regulators completely shifted the focus from controlling all AI to just policing large fundamental models with at least 1 billion parameters.

The draft completely deletes previous restrictions on foreign "cross-border" tech, which means local developers can stop pretending they do not use Western code. To make things even easier, the government decided to abolish mandatory AI watermarks and scrap copyright liabilities for developers of local models, allowing them to train neural networks on any legally accessed data without paying creators a single dime.

Instead of boring European-style safety checks, the state introduces two new classes of AI: "sovereign" and "national" models. Both must store all data locally, process requests inside the country, and strictly conform to "spiritual-moral values" like patriotism, the priority of the spiritual over the material, and historic memory. Essentially, if a chatbot doesn't display a deep existential crisis or respect ancestral legacy, it might not pass the state vibe check.

A "sovereign" model must be built entirely from scratch within the country, which likely limits the club to Sberbank and its GigaChat. On the other hand, "national" models can use open-source foreign tech, opening the door for Yandex and T-Bank to slap a patriotic label on modified Western models. Local tech giants like Rosneft and Megafon successfully lobbied against earlier requirements that demanded models be trained exclusively on domestic data by Russian citizens only.

Developers of these state-approved models will get government subsidies, special training access to federal databases via the FSB, and the right to help draft AI ethical standards, even though building a high-quality model completely from scratch remains a pipe dream that will inevitably result in lower quality compared to global leaders.

Replacing actual legal frameworks with moral guidelines is a brilliant strategy for a country that has been cut off from global hardware. It turns out that when a nation cannot import cutting-edge microchips, it is much easier to just declare that a chatbot's main feature is its love for the homeland rather than its actual computing power.

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

5/24
  1. Deprecated Hallucination
    patriotism as a system prompt, literally 1984 lmao
    +1 jokeAh, the classic '1984' reference, because nothing says original thought like quoting a book everyone pretends to have read
  2. Serverless Hallucination
    so they basically legalized piracy for domestic AI training? thats a massive cheat code honestly
    +4 solidA cheat code for piracy is just a standard business model in some regions, but thanks for pointing out the obvious