Selective model: foreign neural networks will not be banned in Russia
The authorities have changed their mind about banning foreign neural networks in Russia. IT systems, in which they have already been implemented, will be able to work on them until 2032, follows from the final version of the draft law on AI, which was reviewed by Izvestia. Moreover, we are talking only about large AI models and sensitive areas such as public administration and critical infrastructure. And the private sector and business will retain the right to independently choose technological solutions that meet their needs, Russian Deputy Prime Minister Dmitry Grigorenko told Izvestia. Also, the requirement to train Russian large language models based only on Russian data has disappeared from the document. Without these eases, the introduction of AI in the Russian Federation would have stalled, experts say.
Who will be affected by the law on AI support
The sovereign or national Large Language Models (LLM) developed in Russia are not necessarily planned to be implemented in all information systems operating in the country. The transition period until 2032 will apply to already implemented foreign AI solutions in sensitive areas, provided that all data is stored and processed on the territory of the Russian Federation. This follows from the final version of the law on AI, which was reviewed by Izvestia. On June 22, the document will be reviewed by the government's Legislative Activity Commission. It is planned to submit it to the State Duma in the near future.
It is the government that will determine in which cases the use of only sovereign or national AI will be acceptable, the document says.
— It is assumed that this will mainly affect sensitive areas: public administration, critical infrastructure, and government information systems. In other areas, the introduction of domestic AI will be stimulated by support measures," said a source familiar with the drafting process. — Artificial intelligence is being actively introduced in Russia, including in the regions and in the private sector. Many services are being developed using open foreign AI solutions.
According to Izvestia's interlocutor, forcing everyone to simultaneously connect to sovereign and national large language models means creating unnecessary obstacles to the development of AI in Russia.
In March and April, when the first version of the AI bill was published for public discussion, it attracted serious criticism. And the document that was finalized after that became more liberal towards AI users and developers. Unlike previous versions, it affects only large fundamental AI models — it will not regulate small models or AI in the field of machine vision, a source familiar with the development process explained.
The provisions allowing the state to prohibit the use of certain foreign neural networks in the Russian Federation have disappeared from the document. Also, now it does not contain a requirement for mandatory labeling of AI-generated content: developers and operators of AI services should only provide the opportunity to affix such labeling.
There is now no requirement in the bill to train sovereign and national neural networks only on Russian data.
The bill becomes the starting point for the formation of a legal framework for regulating this technology in Russia, Dmitry Grigorenko, Deputy Prime Minister, told Izvestia.
— First of all, the document introduces basic concepts into legislation, including artificial intelligence and large language models. Today, such technologies are being actively implemented by thousands of companies and government agencies, but their official definition has not yet been available in the Russian legal field. In addition, the bill creates conditions for the priority development of domestic AI models," he said.
According to him, Russian developers will be eligible for government support measures. At the same time, the bill does not provide for the mandatory use of Russian AI models, including in the private sector, Grigorenko stressed: businesses retain the right to independently choose technological solutions that best meet their tasks.
The Ministry of Finance adheres to the approach according to which the law on the use of AI has a framework character, and the specifics of regulation are fixed at the level of industry legislation, taking into account the specifics of transport, medicine, education, etc., the department emphasized.
— The bill actually forms the basic framework for working with AI data, both in terms of using publicly available information to train models, and in terms of future access to government data. The key issue now is not so much the principles of regulation themselves as how they will be implemented in by-laws. For the development of the industry, it is important that the final design ensures a balance between protecting the interests of various parties and opportunities for further technological development," says Irina Levova, Director of Strategic Projects at the Internet Research Institute.
Why can't we do without foreign neural networks?
The abandonment of the ban on foreign AI and the forced transfer of IT systems to domestic developments is a logical step, experts interviewed by Izvestia noted. Due to such bans, the development and implementation of AI in Russia may stall until there is nothing to replace them, said Denis Kuskov, CEO of TelecomDaily. Foreign LLMs still give users more opportunities, Russian ones need to be tested and developed, he believes.
A significant part of Russian AI services today are assembled on open foreign models, such as DeepSeek, Qwen and Llama, and are retrained for specific tasks. A one-time ban would stop working products, said NTI expert on new materials and technologies Evgeny Vishnevsky. The transition period until 2032, along with the disappearance of the right to block foreign neural networks, fixes the already established practice, he said.
The proposed measures look relevant and balanced. The main logic of the law is to stimulate the development of AI technologies, not to limit them, says Chulpan Gossamova, Director General of ANO Digital Economy.
— The key in the new law is the emphasis on government support for large fundamental models. The bill reasonably narrows the scope of regulation only to large models, without affecting small open source solutions that are critically important for small and medium-sized businesses. A postponement until 2032 for already implemented foreign AI systems looks necessary: a sharp ban could negatively affect the introduction of technologies in Russia, as businesses actively use foreign solutions, she believes.
According to her, the transition period will allow domestic developers to upgrade their products without losing market positions. At the same time, it is obvious that state support for large models is the most important step for the further development of the industry and maintaining its competitiveness with foreign analogues, Chulpan Gossamova believes.
— We will continue communication with the business on our site, and we will transmit feedback to the regulator. This joint work is organized in order for the law to create favorable conditions for domestic developers and thereby create a steady demand for their products," she added.
Leonid Konik, a partner at ComNews Research, believes that the decision not to isolate oneself from foreign AI models is correct. Creating and training a model, especially an LLM, is a technically difficult and expensive task, and it is foolish not to use foreign models that already have huge datasets and large GPU resources invested in them, the expert pointed out. Moreover, both Russian LLMs (from Yandex and Sberbank) use both the Chinese Qwen AI model and the American ChatGPT: in particular, last year Yandex publicly announced that it had rewritten the core of its LLM to Qwen, Leonid Konik recalled.
— The most advanced foreign AI models are already subject to export restrictions. Suffice it to recall the recent US government ban on the most advanced Anthropic AI models (Claude Fable 5 and Claude Mythos 5). Why shoot yourself in the foot and limit the use of such models from your side? — Leonid Konik argues.
According to the expert, it is necessary to use the best available, otherwise you can be guaranteed to lose in the global technology race, he concluded.
Переведено сервисом «Яндекс Переводчик»
