Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On Radarvc.ru→ original

Roskomnadzor to Spend 2.27 Billion Rubles on AI-Based Traffic Filtering Through 2028

Roskomnadzor has allocated 2.27 billion rubles for a system that will use machine learning to search internet traffic for banned content and ways to bypass blocks. Forbes reports this, citing the digital transformation program for 2026–2028. The model is described as "domestic," and its name has not been disclosed. Experts warn of possible false positives and disruptions to corporate services and VPNs.

AI-processed from vc.ru; edited by Hamidun News
Roskomnadzor to Spend 2.27 Billion Rubles on AI-Based Traffic Filtering Through 2028
Source: vc.ru. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourcevc.ru ↗
Editor-in-Chief
Key metric #1Program period is 2026–2028, budget is 2.27 billion rubles
Key metric #2The model runs on the agency's own infrastructure
Risksfalse blocks and reduced stability of VPNs and business services

Blocking on the Russian internet has so far been built on lists: if an address or a marker is present, access is cut off. Now Roskomnadzor wants to add machine learning to this. According to Forbes, which cites the passport of the digital transformation program for 2026–2028, 2.27 billion rubles has been earmarked for deploying such a mechanism. For anyone who runs services in Russia, builds products for businesses, or depends on VPNs, this is a reason to take a closer look at the resilience of their infrastructure.

What is known about the project

There are few facts, and nearly all of them come from Forbes' retelling. The regulator uses a "domestic" mathematical model whose name is not disclosed. The model is deployed on the agency's own infrastructure, meaning not in a third-party cloud provider's environment. Its task, as the document states, is to automatically extract the relevant information from large volumes of data.

What the open data lacks is detail. It is unknown what data the model was trained on, what share of traffic it will handle, or how its decisions will be reviewed. There are no accuracy figures either. So only the stated goal and the experts' concerns can be assessed, not the real quality of the system's work.

Why the regulator needs this

The logic is easy to explain with a simple example. A classic filter is like a guard with a list of surnames: it recognizes only the people who have already been written down. A trained model is more like an employee who has memorized the typical behavior of violators and notices similar people, even when they are not yet on the list.

According to experts interviewed by Forbes, this is exactly what Roskomnadzor needs: finding banned content and circumvention methods more accurately. The second area stands out in particular. Circumvention tools change quickly, and manually updating lists cannot keep up. Automatic traffic classification theoretically allows faster responses.

What businesses and developers risk

The same experts warn of side effects. The first is false positives. Any model makes mistakes, and in network traffic an error means a legitimate service could fall under blocking or restriction. The second risk is reduced stability of corporate services and VPNs. Many companies use VPNs not to evade restrictions but for everyday work: secure employee access to internal systems, connections between offices, and access to foreign APIs. From the network's perspective, such traffic can look similar to the traffic the regulator wants to stop.

The practical takeaways for teams are these:

  • Plan in advance for disruptions: backup communication channels, alternative routes, and a clear course of action if an external service suddenly stops responding.
  • If your product depends on foreign APIs or VPN connections, add availability checks from Russian networks to your monitoring and keep a history of outages. This will help you separate your own errors from external ones.
  • Do not count on a false positive being corrected quickly. No appeal procedure for such systems has been described so far.

What remains off the record

It is important not to overstate the news. For now, this is a budget line and a statement of purpose, not a report on a working system. According to the publication, the model has already been deployed, but the scale of its use and the timeline for a full launch are unknown. Many of the stated effects are only the experts' opinions.

Still, the sum of 2.27 billion rubles over three years shows that the regulator sees machine learning as a permanent part of its toolkit, not an experiment. For the industry, this means building the possibility of unstable network access into architecture and risk plans. More details will most likely emerge later: as the system launches, businesses and users will see for themselves how it affects the services they rely on.

Story

Key Questions Answered

What is known about the project?

There are few facts, and nearly all of them come from Forbes' retelling. The regulator uses a "domestic" mathematical model whose name is not disclosed. The model is deployed on the agency's own infrastructure, meaning not in a third-party cloud provider's environment. Its task, as the document states, is to automatically extract the relevant information from large volumes of data. What the open data lacks is detail.

Why the regulator needs this?

The logic is easy to explain with a simple example. A classic filter is like a guard with a list of surnames: it recognizes only the people who have already been written down. A trained model is more like an employee who has memorized the typical behavior of violators and notices similar people, even when they are not yet on the list. According to experts interviewed by Forbes, this is exactly what Roskomnadzor needs: finding banned content and circumvention methods more accurately.

What businesses and developers risk?

The same experts warn of side effects. The first is false positives. Any model makes mistakes, and in network traffic an error means a legitimate service could fall under blocking or restriction. The second risk is reduced stability of corporate services and VPNs. Many companies use VPNs not to evade restrictions but for everyday work: secure employee access to internal systems, connections between offices, and access to foreign APIs.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…