Russia’s law on artificial intelligence is Federal Law No. 243-FZ of 26 July 2026 “On Supporting the Development of Artificial Intelligence Technologies in the Russian Federation”. It has applied since 1 September 2026; some of its provisions take effect on 1 March 2027. The law does not regulate all of AI, only large foundation models of 1 billion parameters or more. It does not deal separately with AI agents or with liability for what they do.
The quotations and article titles below are our translation; the official text is in Russian.
In brief
- Subject: large foundation models. The law regulates the development, deployment and use of AI models that contain at least 1 billion parameters and serve as the basis for other software (Article 1; Article 3, point 2).
- A developer is also whoever modifies a model. A company that fine-tunes someone else’s model may become a developer within the meaning of the law (Article 3, point 5).
- Sovereign and national models. From 1 March 2027, the Government will set the cases in which only such models are allowed; systems already in operation get a transition period until 1 September 2032, provided data are processed and stored in Russia.
- Labelling. From 1 March 2027, there is an option to mark audio and video materials as made with AI, and large platforms (more than 500,000 users a day) must give users that option.
- Liability: under general law. There is no special regime for harm caused by AI (Article 11).
What is the law called, and when was it adopted?
Federal Law No. 243-FZ of 26 July 2026 “On Supporting the Development of Artificial Intelligence Technologies in the Russian Federation”. It was adopted by the State Duma on 8 July 2026, approved by the Federation Council on 17 July 2026 and signed by the President on 26 July 2026. The law has 13 articles.
From what date do its provisions apply?
Under Article 13, the law enters into force on 1 September 2026, except for certain provisions:
| Provisions | From |
|---|---|
| The law as a whole: definitions, principles, powers, support measures, liability | 1 September 2026 |
| Article 5, part 2, points 3–5: cases where only sovereign and national models may be used; sector requirements for preventing risks; access to data of state systems for training | 1 March 2027 |
| Article 6, parts 2–5: what a sovereign and a national model are | 1 March 2027 |
| Article 8: obligations of developers of sovereign and national models | 1 March 2027 |
| Article 9: labelling of materials created with models | 1 March 2027 |
| Article 10: use of intellectual property in training | 1 March 2027 |
Transition rule (Article 13, part 3): until 1 September 2032, the requirement to use only sovereign or national models does not apply to systems in which models were created or are operated on the day the requirement enters into force, provided that data are processed and stored in Russia.
What does the law require from companies that use AI?
The law sets few direct obligations for a company that simply uses a ready-made model. Four things matter:
- Fine-tuning can make a company a developer. Under Article 3, point 5, a developer is a person who develops or modifies a large foundation model. A company that fine-tunes an open model may become a developer.
- Watch for the Government resolution on sovereign models. From 1 March 2027, the Government may set the cases in which only sovereign or national models are allowed (in the financial sector, in agreement with the Bank of Russia) and establish sector requirements for preventing risks.
- Labelling. A person who creates audio and visual materials with a model is given the option of placing a notice that AI was used; the format is set by agreement with the model’s provider. Large platforms with a daily audience of more than 500,000 users in Russia must give users that option.
- If a company develops a sovereign or national model, from 1 March 2027 it must take organizational and technical measures to secure the model, set rules for its operation and keep technical documentation (Article 8).
What are a sovereign and a national model?
Both are models of a Russian legal entity whose answers to users are prepared, and whose data are stored, in data centres in Russia owned by Russian legal entities.
- Sovereign (Article 6, part 2): the Russian developer defines and changes the model’s characteristics at every stage of its life cycle, the development cycle including training is fully technically reproducible, and the model has passed conformity assessment in the procedure set by the Government.
- National (Article 6, part 3): the essential characteristics are defined by the Russian developer, and the components, including other parties’ models, are distributed under an open licence.
Does the law cover AI agents and liability for their actions?
No. The word “agent” does not appear in the law, and there are no special rules on harm caused by the actions of AI. Article 11 refers to general law: participants in these relations are liable “in accordance with the legislation of the Russian Federation”. The law gives the developer of a sovereign or national model the right to insure liability for harm caused when the model is used (Article 7, point 5).
For a company, this means that liability for an agent’s actions follows the general rules of civil law: the company answers to its counterparties, and its chief executive answers to the company (Article 53.1 of the Civil Code of the Russian Federation on compensation of losses). The AI law does not change this. More on this on the page who is liable for AI mistakes.
Among the law’s principles is a risk-based approach: requirements for the use of models are set according to the likelihood and scale of harm to life, health, business reputation and property (Article 4, point 5). The Government may set specific sector requirements from 1 March 2027.
How does the law relate to FSTEC Order No. 117?
They are different levels. Law No. 243-FZ sets the framework for large foundation models. FSTEC Order No. 117 sets requirements for protecting information in state information systems. The draft amendments to Order No. 117 of 24 August 2026 rely on the definition of artificial intelligence in Article 3, point 1 of Law No. 243-FZ and introduce measures for AI models and agents in state systems from 1 March 2027. More on the page about Order No. 117 and AI agents.
What to do
- List the models the company uses, and mark which of them are large foundation models (1 billion parameters or more).
- Check whether the company has become a developer by fine-tuning or modifying a model.
- Follow the Government resolutions that take effect on 1 March 2027: on the cases where only sovereign and national models may be used, and on sector requirements.
- For agents, the law creates no new obligations but removes none of the old ones: liability for their actions follows the general rules. So you need rules for what an agent may do, and a log that proves those rules were followed.
- For state systems, check the requirements of the draft amendments to FSTEC Order No. 117.
Frequently asked questions
Does Russia have a law on artificial intelligence?
Yes. Federal Law No. 243-FZ of 26 July 2026 “On Supporting the Development of Artificial Intelligence Technologies in the Russian Federation” has applied since 1 September 2026; some provisions apply from 1 March 2027. It regulates large foundation AI models.
What does Law No. 243-FZ regulate?
The development, deployment and use of large foundation AI models, that is, models of 1 billion parameters or more that serve as the basis for other software. The law introduces the notions of a sovereign and a national model, support measures, obligations of their developers and rules for labelling materials created with AI.
Does Law No. 243-FZ affect a company that simply uses ChatGPT or another model?
Directly, only a little. But if a company fine-tunes or modifies a model, it may become a developer (Article 3, point 5). From 1 March 2027, the Government may set the cases in which only sovereign or national models are allowed; systems already in operation on that day get a transition period until 1 September 2032, provided data are stored in Russia.
Who is liable for harm caused by AI under Law No. 243-FZ?
The law introduces no special rules: under Article 11, liability arises in accordance with the legislation of the Russian Federation, that is, under general law. The developer of a sovereign or national model may insure liability for harm caused when the model is used (Article 7, point 5).
Does AI-generated content have to be labelled?
From 1 March 2027, the law provides the option of placing a notice that AI was used in audio and video materials; the format is set by agreement with the model’s provider. Large platforms with a daily audience of more than 500,000 users in Russia must give users that option (Article 9).
Sources
- Federal Law No. 243-FZ of 26 July 2026 “On Supporting the Development of Artificial Intelligence Technologies in the Russian Federation” (in Russian), ConsultantPlus: https://www.consultant.ru/document/cons_doc_LAW_540336/
- Official Internet Portal of Legal Information (in Russian): http://publication.pravo.gov.ru
- Draft amendments to FSTEC of Russia Order No. 117, regulation.gov.ru, project 170500 (in Russian): https://regulation.gov.ru/projects/170500
Checked: 16 October 2026. The legal conclusion in a specific situation belongs to the organization’s lawyer.