During the XIV St. Petersburg International Legal Forum (SILF-2026), the Ministry of Justice of the Russian Federation proposed the introduction of strict filtering mechanisms for domestic artificial intelligence systems. Deputy Minister of Justice Oleg Sviridenko stated that major technology companies are obligated to integrate software filters aimed at protecting sovereignty and national security.
A key vector of the proposed regulation is the fight against the dissemination of information from individuals recognized as foreign agents. The ministry is concerned that modern large language models (LLMs) are capable of automatically reproducing, compiling, and scaling the theses outlined in materials by such individuals. In this regard, the Ministry of Justice proposes to oblige owners of AI platforms, including YandexGPT and GigaChat services, to completely exclude such data from training datasets and user search results.
Developer Liability and Foreign Experience
Representatives of the ministry propose enshrining in law the principle that the company developing or operating the software bears responsibility for the content generated by the neural network. As measures of impact, the introduction of administrative, criminal, and financial liability is being considered. The Ministry of Justice explicitly stated that sanctions could be applied by analogy with measures used against foreign IT corporations for violating moderation rules.
As an example of state control over generative algorithms, the regulatory model of the People's Republic of China was cited. In the PRC, providers of generative AI bear direct responsibility for the compliance of system responses with official regulatory safety standards.
Technological Challenges and Industry Position
Representatives of the technology sector and relevant experts, in particular analysts from the ComNews group of companies, note that the practical implementation of the Ministry of Justice's requirements is fraught with difficulties. The mechanical exclusion of direct publications from foreign agent resources is technically feasible. However, identifying retellings, citations, interpretations, or mentions of the same concepts in third-party sources requires significant complication of pre-moderation algorithms.
Experts warn that excessive filtering may negatively affect the overall variability and relevance of model responses, especially in humanities knowledge segments. At present, leading Russian AI developers carry out content moderation within the framework of the signed Code of Ethics in the field of artificial intelligence. Existing models already contain basic restrictions on the generation of destructive content and instructions for violating legislation.
The introduction of the measures proposed by the Ministry of Justice will require the development of detailed subordinate acts and verification by the Ministry of Digital Development and Roskomnadzor. The main task of regulators is to find a balance between national security requirements and maintaining the pace of technological development of domestic IT products.