Materials for the draft national plan for the introduction of artificial intelligence until 2030 revealed a section dedicated to the automated analysis of VPN service features. According to the document, Russian authorities considered using neural networks to bring the effective blocking level of virtual private networks up to 95%. Neural networks were supposed to analyze signs of up to 70% of the total number of blocked services, which would significantly expand the scale of automated traffic filtering compared to current methods.
What Was Recorded in the Draft National Plan
The draft edition of the document provided a clear quantitative target: 95% effectiveness in blocking VPN services using AI technologies. The scope of neural network application covered up to 70% of all blocked services, which would mean a transition from manual and semi-automated analysis to predominantly machine-based analysis. Such an approach would allow a prompt response to the emergence of new bypass protocols and reduce dependence on regular updates of banned address lists.
Ministry of Digital Development: The Measure Is Absent in the Current Edition
The Ministry of Digital Development, Communications and Mass Media of the Russian Federation stated that this measure is absent in the current edition of the national AI plan. The agency clarified that currently, "intelligent algorithms" are used to analyze network activity, which do not constitute full-fledged artificial intelligence. Thus, significant changes occurred between the draft and final versions of the document regarding the application of neural networks to the task of blocking VPNs.
TSP Infrastructure Continues to Scale
Despite the absence of the AI line in the final national plan, Roskomnadzor continues a large-scale update of the TSP infrastructure (technical means to counter threats). The throughput capacity of the ASBI (automated blocking and identification system) is planned to increase to 954 Tbps by 2030. Earlier, Forbes reported the allocation of 2.27 billion rubles for the introduction of machine learning tools for traffic filtering, which indirectly confirms the regulator's ongoing interest in automated methods of network stream analysis.
Experts: Detection Accuracy Drops to 38% with Masking
Network market specialists note that AI is indeed capable of detecting VPN traffic by indirect signs: packet transmission rhythm, session duration, hosting characteristics, and connection patterns. However, at the current stage of technology development, detection accuracy drops sharply when header masking mechanisms are used — according to experts, it falls to 38%. This means that even with the full implementation of neural network filtering, a significant portion of traffic disguised as ordinary HTTPS or using obfuscation will remain invisible to the system.
Risks for Legitimate Traffic and Corporate Infrastructure
A separate concern among experts is the problem of massive false positives. Under aggressive detection parameters, the AI system risks blocking legitimate corporate traffic, cloud services, and IP telephony, which in their network characteristics may intersect with the signs of VPN connections. In conditions where a significant portion of business processes in Russia is tied to cloud infrastructure and remote access, such failures can cause direct economic damage to companies and state organizations.
