The issue of labeling content created using large language models has once again come into focus for Russian lawmakers. Based on recent initiatives, labeling images, audio, and video generated by artificial intelligence may become not voluntary but strictly mandatory. This concerns amendments to the recently adopted law on AI and the Civil Code, which are currently being actively developed at the expert level.
The Essence of the Legislative Initiative
A working group under the Council for the Codification of Civil Legislation has prepared a draft that radically changes the approach to regulation. While the current version of the law only recommends notifying about the use of neural networks, the new initiative introduces a strict requirement. At the same time, the wording "created using" raises many questions: even materials merely edited by AI could fall under it, without a clear indication of the minimum share of machine intervention.
The most problematic areas are the lack of clarity regarding who exactly should place the label: the content author or the platform distributing it. It is also not defined whether the label should be preserved when transferring files between different services and who will monitor compliance with the rule. Large platforms with a daily audience of more than 500,000 users are already required to technically ensure the ability to place such a label, but the specific authorized body has not yet been named.
Criticism and Risks
In the professional community, the initiative faces serious resistance. My analysis shows that the main argument against it is the ineffectiveness of combating deepfakes. Creators of malicious content will easily circumvent the requirement by using foreign services that do not fall under Russian jurisdiction and distributing materials bypassing the labeling. Moreover, the rule could create an illusion of control, while real risks only worsen.
Technological unpreparedness is another stumbling block. The global community has still not reached a unified standard for preserving machine-readable labels during further manipulations with content. Premature implementation of mandatory requirements may not only fail to solve the problem but also slow down the development of domestic AI services, creating unnecessary bureaucratic barriers.
International Context and Prospects
Interestingly, there are already benchmarks to follow in the world. The Chinese approach, where services add visible and hidden labels and platforms verify them, looks more holistic. The European AI Act also requires providers to implement machine-readable labeling and distributors to disclose the origin of deepfakes. It is quite likely that Russia will ultimately arrive at a mixed model, combining mandatory requirements for major players and recommendatory ones for the rest.
For now, the draft is at the discussion stage, and the final form of the rule will depend on which option the drafters of the amendments choose. Given the complexity and ambiguity of the issue, I expect that the final version will be significantly refined. However, the very fact of moving toward mandatory labeling is an important signal for the market: AI regulation in Russia is becoming more mature and systematic, which in the long term could create a more transparent and secure digital environment.