Roskomnadzor deployed a traffic filtering mechanism based on machine learning algorithms; 2.27 billion rubles were allocated for its implementation. The solution is integrated into the automated security system for the Russian segment of the Internet (ASBI) and complements the work of technical means of countering threats (TSPU). The new technology will help more effectively identify content and ways to bypass blocking. Experts do not rule out false positives and decreased stability of VPN and corporate services.
Image source: ELLA DON / unsplash.com
The creation and implementation of traffic filtering mechanisms based on machine learning algorithms is mentioned in the passport of the digital information program for 2026, as well as for the planning period of 2027 and 2028. Roskomnadzor spent 2.27 billion rubles on the new system. It is based on a domestic artificial intelligence model, which the department deployed in its own infrastructure. This year, operators have allocated 3.99 billion rubles for the installation of new and modernization of old TSPU. from the main one and 6.2 billion rubles. from additional funding.
The intention to create and implement AI-based filtering tools was announced in January. In May it became known that for this purpose it is planned to use the Astra AI neural network created by the Astra Group. According to experts, the system does not use a large language model, but a predictive model: it does not analyze the content of traffic, but estimates the likelihood of using a VPN based on external signs. This means that the quality of VPN services may deteriorate, even if they mask traffic and change addresses. Legitimate corporate services may also be blocked.
The AI implemented in the TSPU will help detect not only means of bypassing blocking, but also sites and other prohibited resources on which prohibited materials are published. At the same time, AI will not provide 100% accuracy—the likelihood of errors and false positives remains. In addition, with modern connection speeds, AI threatens to become a bottleneck – a source of delays and user irritation.
A possible function of AI is to analyze and accelerate the development of signatures that are used by DPI systems to recognize and filter traffic without decrypting its contents. DPI systems rely heavily on signatures being up-to-date, otherwise filtering efficiency is reduced. Owners of blocked sites often create mirrors of them on the Internet, and prohibited content may appear on other resources. It is impossible to manually identify such a number of sites and materials. Perhaps machine learning will help detect such materials by checking against a sample, and not just by domain name.
If you notice an error, select it with the mouse and press CTRL+ENTER.
