The Russian internet has entered the era of «algorithmic censorship». Roskomnadzor ended 2025 with record blocking figures and announced the implementation of artificial intelligence for total traffic control. The state is moving from targeted content removal to a systemic cleansing of the anonymity infrastructure.
1. Isolation Statistics: 1235% vs. Freedom of Access
The agency's report for the past year demonstrates an unprecedented rise in repressive activity. The total volume of blocked materials reached 1.289 million units, which is 59% higher than the figures for 2024. However, the main figure of the report lies in the details: the number of blocked materials regarding methods to bypass blocking (VPN, proxies) skyrocketed by 1235%, exceeding 93,000 units.
authorities are no longer limiting themselves to fighting political content or fake news. The priority has become the destruction of tools that allow users to maintain access to free information. LGBT content also came under attack (a 269% increase in blocks) along with other categories, but the war on VPNs became the main trend of the year.
2. «Oculus» and «Vepr»: The Technological Contour
Experts link the growth in blocking efficiency to the full-scale deployment of automated surveillance systems. Roskomnadzor now relies on a triad of technologies:
- IS «Mir»: A basic search engine for banned content in texts.
- IS «Oculus»: A computer vision system that automatically finds violations in images and videos, replacing hundreds of moderators.
- IS «Vepr»: An analytical platform for predicting «information bombs» and identifying tension points in the network.
3. Plan for 2026: AI vs. Encryption
This year, the regulator intends to take the next step. According to the digitalization plan, Roskomnadzor has allocated 2.27 billion rubles to create a traffic filtration system based on machine learning (ML).
The new algorithm must solve two fundamental tasks that were previously beyond the power of ordinary blocking. First, semantic analysis of «mirrors»: AI will find copies of banned sites not by IP address, but by the meaning of the text and page structure. Second, behavioral analysis of encrypted traffic is being introduced. The system will learn to identify VPN connections and messengers even inside encrypted channels by analyzing not the content of packets, but their metadata and behavior.
