The State Duma has passed an AI regulation law in its third reading. The document introduces the very concept of AI into Russian law for the first time and requires that sovereign and national models be stored exclusively on Russian territory. Let’s break down who this actually affects in practice and whether we should worry about access to the familiar neural networks that most of us use today, such as ChatGPT, Claude Code, DeepSeek, and others.
What will happen to AI in Russia now?
What AI Law Did the State Duma Pass
The AI law does two things:
- First, it provides a legal definition of artificial intelligence — previously, Russian law simply had no such concept.
- Second, it introduces requirements for where data is processed for specific categories of models.
The law also introduces the concept of a large foundational model — a large neural network trained on massive volumes of data, containing at least 1 billion parameters (internal settings that affect its responses), and capable of solving many different tasks. In simple terms, this refers to systems like YandexGPT, GigaChat, or GPT from OpenAI.
An important point: the law describes principles, not ready-made rules. The specifics — a list of sensitive areas, data access procedures, and developer support measures — will be detailed by the government after the law is signed by the president.
Sovereign and National AI Models — What’s the Difference
The main detail that scared everyone with the phrase “store only in Russia” concerns specifically sovereign and national models. These are not just any neural networks, but Russian models created with the participation of Russian legal entities.
For them, the law introduces two requirements:
- data processing must take place on Russian territory
- the government commits to supporting developers, including access to training data
The localization requirement applies only to Russian sovereign and national models
In other words, this is about developing domestic neural networks and where their servers and data are physically located. For foreign services like ChatGPT or Claude, the localization requirement in this part does not directly apply — they simply don’t fall under the definition of a national model.
Which Neural Networks Will Continue to Work in Russia
The most common question is whether ChatGPT, Claude, and DeepSeek will be blocked. According to the text of the law, there is no direct ban on foreign neural networks. The document regulates where Russian models store their data and sets a framework for future rules, but does not introduce blocks on specific foreign services.
This means that Russian neural networks (YandexGPT, GigaChat) will continue to operate and receive government support, while access to foreign services is determined more by their own policies and general restrictions than by this law.
There is no direct ban on foreign neural networks in the law
Important distinction: ChatGPT’s availability for Russians today is primarily limited by OpenAI’s own decisions, not by this law. The new document neither resolves nor worsens this issue.
When Will the AI Law Take Effect and What Will Change
After being passed by the State Duma, the law must go through the Federation Council and receive the president’s signature. Only after that will the government begin developing specific rules for AI application — these will contain the details that businesses and developers will actually feel.
What the government still needs to determine:
- a list of sensitive areas where AI use will be more strictly regulated
- procedures for accessing data for model training
- mechanisms for government support of developers
Until these secondary regulations are adopted, nothing changes for the average user. No service is being shut down because of the law itself — it’s a framework document that sets a direction, not a kill switch.
How the AI Law Will Affect the Average User
In short — right now, it won’t. You continue to use the same neural networks as before, and the changes will mainly affect companies that develop and train their own models in Russia.
For the average user, nothing changes for now
For Apple device owners, there’s no reason to panic either. Apple Intelligence doesn’t officially work in Russia, but that’s due to Apple’s own policy, not the new law. Meanwhile, the company is already preparing flexibility in another direction — according to rumors, a new feature in iOS 27 will allow users to replace the built-in AI with ChatGPT, Gemini, or Claude.
It’s worth keeping an eye on this topic if you’re building a business on neural networks or professionally working with data processing — the upcoming government rules will be relevant to you. Everyone else can calmly continue using their usual services: the AI law is a foundation for regulation in the years ahead, not a ban that will be switched on tomorrow.
