The National AI Adoption Plan for 2027–2030 is a government document that turns scattered artificial intelligence projects into a single program: more than 170 metrics across 16 industries, 32 responsible agencies, support measures, and the removal of administrative barriers. There are no direct obligations for private companies in the details released so far, but the plan sets the direction: agencies will encourage businesses in their sectors, while the government is also preparing a transition to domestic models for critical infrastructure. Companies that already know which models they use, where data is processed, and what impact AI delivers will move through this period more smoothly than others.
Prepared by AIrassvet from publicly available sources, verified October 10, 2026. It is intended for business owners, CEOs, and CIOs of Russian small and midsize companies, as well as executives responsible for AI in the industries included in the plan. We cover only what official officials and business media have reported; the full text of the plan was not publicly available at the time of verification. This article is not legal advice. AIrassvet is not affiliated with the agencies, companies, or publications mentioned, and the section at the end describes the publisher’s services.
Contents
- What happened in early October 2026
- Quick take: the key points of the national plan
- What is known about the plan, and what is not yet known
- How the plan relates to Federal Law 243-FZ and the shift to domestic AI
- Which industries it will affect first
- Where Russian businesses stand now
- Company preparation checklist
- 90-day plan
- When there is no need to rush
- Common mistakes
- Frequently asked questions
- Primary sources and verification method
- How AIrassvet helps companies prepare for the national plan
- Bottom line: prepare for measurable results, not reports
What happened in early October 2026
Three news items appeared in one week, together explaining why this topic became so hot.
October 2 Kommersant reported details of the draft National Plan for AI deployment in the sectors of the economy, social services, and public administration for 2027–2030. According to the publication, the draft was approved by the profile subcommittee of the government chaired by Deputy Prime Minister Dmitry Grigorenko, and the Deputy Prime Minister’s office confirmed this. The next step is review by the presidential commission on AI technology development.
October 6 Prime Minister Mikhail Mishustin at the Digital Solutions forum said, that the rollout of mandatory domestic AI will take place with a transition period, and authorities will determine when the use of only sovereign models will be possible if critical infrastructure is involved. According to him, the transition period is needed “so that enterprises and companies have time to prepare.”
October 8 Dmitry Grigorenko told Vesti that the plan is expected to be adopted by the end of October. According to him, the document includes a list of measures with yearly metrics, expected outcomes, and impact assessments.
For businesses, this is a signal: AI adoption is moving from a phase of voluntary experiments to a state program with sector-specific targets and results monitoring.
Quick take: the key points of the national plan
| Parameter | What is known as of October 10, 2026 | Source |
|---|---|---|
| Effective period | 2027–2030 | Kommersant, 10/02/2026 |
| Status | Approved by a government subcommittee, awaiting the presidential commission; adoption expected by the end of October | Kommersant, 10/02/2026; Vedomosti, 10/08/2026 |
| Scale | More than 170 federal project metrics across 16 industries | Kommersant, 10/02/2026 |
| Responsible parties | 32 federal agencies | Kommersant, 10/02/2026 |
| Structure | Three blocks: adoption metrics, reduction of administrative barriers, and support measures for development and implementation | Kommersant, 10/02/2026 |
| Regions | Metrics are divided into mandatory and voluntary; monitoring is in real time online | Kommersant, 10/02/2026 |
| Domestic AI | For critical infrastructure, sovereign models with a transition period | Vedomosti, 10/06/2026 |
One discrepancy is worth keeping in mind. Kommersant writes about more than metrics, while Vedomosti, in its retelling of Grigorenko’s interview, refers to roughly 170 measures. Since the text of the plan has not been published, we consider the wording “about 170 items with yearly metrics” to be the most accurate and do not draw conclusions based on the exact count.
What is known about the plan, and what is not yet known
What is known
- Agencies are responsible for two things. According to Kommersant, federal bodies will ensure the use of AI both in their own work and in the industries they oversee, “encouraging business to achieve one indicator or another.”
- The plan builds on obligations already in place. Right now, the obligation to implement AI at the agency and industry level is set out in digital transformation programs for 2026; this experience is planned to be scaled up for 2027–2030.
- Regions will participate in different ways. Mandatory metrics apply where a federal target cannot be achieved without the region; the rest are voluntary. After mandatory plans are completed, regions will be able to connect to centralized federal solutions.
- The plan is expected to be evaluated by impact. Both the AI Alliance and the experts quoted by the publication frame the main challenge the same way: move from tracking activity to measuring outcomes.
What is not yet known
- the specific metrics for each of the 16 industries;
- the funding amount and the criteria for support measures;
- which administrative barriers are planned to be removed, and by when;
- whether any requirements apply directly to private companies outside critical infrastructure;
- a list of cases where only sovereign models are allowed.
It makes sense to postpone any decisions that depend on these points until the document is published. But the preparatory work from the next sections can start now: it will be useful no matter what the final text says.
How the plan relates to Federal Law 243-FZ and the shift to domestic AI
The national plan is not a law, but a government action program. The legal framework for models is set by Federal Law No. 243-FZ of 07/26/2026 “On Supporting the Development of Artificial Intelligence Technologies in the Russian Federation.” The law regulates large foundational models, and the key business dates look like this:
| Date | What happens | What it means for the company |
|---|---|---|
| September 1, 2026 | 243-FZ takes effect | The concepts of a sovereign and national model are introduced |
| October 2026 | The national plan for 2027–2030 is expected to be adopted | Sector-specific goals and support measures will appear |
| March 1, 2027 | The government determines the cases in which only sovereign or national models are allowed | Check whether your AI systems fall into such cases |
| Through September 1, 2032 | Transition period for systems that are already operating, provided data is processed and stored in Russia | The location of data processing becomes a decisive factor |
Mishustin’s comments about a transition period for critically important infrastructure are consistent with this framework. The term “critical information infrastructure” is defined by Federal Law No. 187-FZ, and if your company or your customer is a CII entity, the requirements for sovereign models will likely affect you first. For a detailed breakdown of the law’s articles, see the article “AI Law 243-FZ: What It Means for Business”.
Which industries will be affected first
Kommersant names public administration, construction and utilities, the fuel and energy sector, manufacturing, retail, and transportation among 16 industries. The full list has not been published. Below is our editorial assessment of where companies should expect changes, based on the logic that “the ministry is responsible for the industry metric.” This is not a prediction of the plan’s contents.
| Industry on the list | Where demand for AI is likely | What to prepare |
|---|---|---|
| Public administration | Government contractors, handling citizen requests, document workflows | Requirements for data processing location and domestic models in the statement of work |
| Construction and utilities | Review of design documentation, estimates, dispatch services for management companies | Baseline for timelines and errors before rollout |
| Fuel and energy sector | Predictive maintenance, consumption forecasting | CII status and an environment without external APIs |
| Manufacturing | Quality control, computer vision, demand forecasting | Data and metrics on defects, downtime, and inventory |
| Retail | Demand forecasting, pricing, customer support | Tracking the impact on revenue and margin, not the number of bots |
| Transportation | Shipping documents, planning, dispatching | Integrations with accounting systems and an audit trail of decisions |
If your company operates in one of these industries or sells solutions to such customers, industry-specific requirements, support programs, or requests for impact reporting may come to you through the relevant ministry.
Where Russian business stands now
Recent surveys show that AI is already in companies, but it is becoming manageable more slowly than it is spreading.
- Large businesses are actively experimenting. Vitaly Titov, Director of the consulting firm Yakov and Partners, told RIA Novostithat in a survey of more than 150 large Russian companies, over 70% had integrated generative AI into at least one business process.
- Domestic models are still in the minority. According to an Arenadata survey of participants in the Digital Solutions forum, 76% of organizations use AI, about two-thirds work with American LLMs, and the share of companies using domestic models does not exceed 20%. Qwen and ChatGPT lead the pack, used by roughly half of respondents.
- Maturity is low. In the same survey, end-to-end process automation was implemented by 12% of participants, agentic workflows were deployed by 10%, and 78% of organizations have no working AI usage policy.
The Arenadata sample size was not disclosed in the CNews publication, and the Yakov and Partners survey covers only large companies, so the figures describe a trend, not the entire market. But the takeaway for preparing for the national plan is the same: most companies do not have a complete list of which models they use and where. Without that list, it is impossible to assess what will change when moving to sovereign models.
Company prep checklist
We selected six steps that are useful under any final version of the plan: they are based on the already effective 243-FZ, the stated transition period, and the stated principle of impact assessment. We deliberately do not include steps that depend on unpublished metrics.
- Create an inventory of AI systems and models. For each use case, record the model and vendor, the country where requests are processed, the type of data, and the process owner. A template for the inventory is available in the article on the company AI usage policy.
- Determine CII status. Check with security and legal counsel whether you or your key customer are a subject of critical information infrastructure. This determines how urgent the move to sovereign models will be.
- Make the model replaceable. If a scenario is tightly tied to a single foreign API, the transition becomes a ground-up project. A routing layer between the application and the model lets you switch models without rewriting the process; we described how this works in the article on an AI model router.
- Test domestic and open models on your own tasks. Compare not rankings, but quality on 50–100 real examples from your process. An overview of options is available in the article “LLMs in Russia 2026: GigaChat, YandexGPT, Qwen, DeepSeek”, and for hosting open models, see the article on inference for open-source LLMs in a Russian cloud.
- Establish the baseline for the impact. If the plan evaluates results rather than the number of projects, companies with “before” and “after” measurements will be in a stronger position both for reporting and for funding applications. For each scenario, record current time, cost, error rate, and transaction volume.
- Track available support measures. The third section of the plan is devoted to support for AI development and deployment. While the criteria are still unknown, it is useful to know the mechanisms already in place — they are covered in the article on grants for AI projects.
Why this approach
We prioritize the registry and model replaceability over choosing a specific domestic vendor for a simple reason: the government will only define the list of cases for sovereign models by March 1, 2027, and choosing a vendor today may be premature. A registry and an architecture with a replaceable model will be useful under any outcome. The cost of this approach is extra work at the start and a small layer of infrastructure, but that is less than an emergency migration.
90-Day Plan
| Period | Actions | Result |
|---|---|---|
| Weeks 1–2 | Inventory of AI scenarios: survey of departments, subscription invoices, access rights | A registry with the model, processing country, and owner for each scenario |
| Weeks 3–4 | Data classification and CII status with legal and security teams | A list of scenarios where a move to the Russian environment is needed first |
| Weeks 5–8 | Pilot an alternative model on one process through a routing layer | Comparison of quality and cost on real-world examples |
| Weeks 9–10 | Baseline impact metrics for 2–3 key scenarios | “Before” metrics suitable for reporting and applications |
| Weeks 11–13 | Review the published plan and industry metrics | An updated roadmap for 2027 |
If the plan is adopted by the end of October, you will already have a document ready for comparison by the final stage. If adoption is delayed, the first ten weeks will still give you a controlled environment and a clear migration cost.
When there is no need to rush
Not every company needs to make an urgent change:
- AI is used only with public data. If employees generate marketing copy without personal or confidential data, the transition period and industry metrics will probably have little effect on you. A registry and a policy will be enough.
- You do not work with CII or government customers. Mishustin’s statement applies to critically important infrastructure. For the rest of business, the reference point is the dates in Federal Law No. 243-FZ and the final text of the plan, not an emergency migration.
- The scenario is still experimental. If AI is not embedded in the process and does not affect customers, it is easier to wait for the requirements to be published than to migrate a prototype.
Common mistakes
- Waiting for the plan text and doing nothing. Inventory and baseline impact measurements are needed under any scenario and take weeks, not days.
- Migrating everything at once. A broad switch without comparing quality on your own data often makes results worse. Start with one process.
- Reporting the number of bots. If the plan evaluates impact, the number of launched services without metrics proves nothing. Experts in a Kommersant article speak directly about this risk.
- Confusing the national plan with the law. The national plan sets goals and support measures, while mandatory rules for models are established by Federal Law No. 243-FZ and government decrees.
- Relying on retellings. The figures already differ across publications. Check the original source after publication.
Frequently Asked Questions
What is the national AI implementation plan?
It is a government document for 2027–2030 that combines separate federal, industry, and regional AI initiatives into one program. According to Kommersant, it includes more than 170 metrics across 16 industries overseen by 32 agencies, as well as sections on removing administrative barriers and support measures.
When will the national AI plan be adopted?
Deputy Prime Minister Dmitry Grigorenko said on October 8, 2026, that the plan is expected to be adopted by the end of October. As of October 10, the document has been approved by the government subcommittee and is awaiting review by the presidential commission on AI technology development.
Is business required to meet the national plan’s targets?
In the details published so far, the obligations fall on federal agencies and, in part, on the regions. Agencies are supposed to encourage businesses in their sectors, but no direct requirements for private companies have been described in open sources. That will become clear only after the text is published.
Do companies need to switch to domestic AI models?
Mishustin spoke about transitioning critical infrastructure to sovereign models during a transition period. Under Federal Law No. 243-FZ, the cases in which only sovereign or national models are allowed will be defined by the government starting March 1, 2027, and for existing systems the transition period runs until September 1, 2032, provided data is processed and stored in Russia.
What should a company do right now?
Create a registry of AI scenarios showing the model and data processing country, clarify CII status, make the model replaceable through a routing layer, test alternative models on one process, and record the baseline impact.
Will there be support measures for companies implementing AI?
Yes, one of the three sections of the plan is devoted to support measures for AI development and deployment. The criteria and funding amount have not yet been published.
Primary sources and verification method
We collected publications from October 2–10, 2026, saved the page texts, and compared wording across outlets. Each fact is labeled with a source and date; discrepancies (“metrics” and “activities”) are noted in the text. Industry estimates in the table are marked as editorial and are not presented as plan content.
- Kommersant, 02.10.2026: project plan details and news about subcommittee approval.
- Vedomosti, 06.10.2026: Mishustin on the transition period.
- Vedomosti, 08.10.2026: Grigorenko on the adoption timeline.
- Inc., 02.10.2026: on a unified system of metrics.
- CNews, 09.10.2026: Arenadata survey; sample size not disclosed.
- RIA Novosti, 10/06/2026: data from Yakov and Partners, more than 150 large companies.
- ConsultantPlus: 243-FZ.
Limitations: the plan text was not published as of the review date; the sample size and methodology for some surveys were not disclosed; data on small businesses is absent from the surveys found.
How AIrassvet helps you prepare for the national AI plan
Preparing for the national AI plan and the shift to domestic models is an engineering task: you need to understand which models work where and make sure they can be replaced without stopping operations. On the AIrassvet website, AIrassvet website the following areas are described that relate directly to this:
- Auditing processes where AI is already in use: a register of use cases, models, data, and metrics.
- Deployment and fine-tuning of language models on company servers when data must remain within the Russian data environment.
- Integration and testing of an alternative model in the workflow, including quality comparison on your data.
- Training the team to work with the new environment and post-launch support.
The first step is to choose one process where AI is already running on a foreign model, describe its current performance metrics, data sources, constraints, and the criterion by which you will consider the replacement successful.
Bottom line: prepare for measurable results, not reports
The national AI implementation plan has not yet been published, but its outline is already clear: industry-specific targets, agency accountability, support measures, and an assessment of real-world impact. At the same time, 243-FZ and Mishustin's statement set the course toward sovereign models for critical infrastructure, with a transition period.
The most reliable preparation does not depend on the details of the plan: an inventory of AI systems, a clearly defined data processing location, a replaceable model, and before-and-after measurements for at least two or three processes. Start with an inventory — once the plan is published, you will be able to compare it with your own map instead of building it in a rush.