A tectonic shift is brewing in the Russian legal landscape: labeling images, audio, and video created using large language models may cease to be voluntary. This concerns amendments to the recently adopted AI law and the Civil Code, which are currently being actively developed within the working group of the Presidential Council of the Russian Federation.
The Essence of the Legislative Initiative
The draft, prepared under the leadership of Pavel Krasheninnikov, Chairman of the State Duma Committee on State Construction, proposes a radical change in approach. The current version of the law only recommends marking synthetic content, while the new initiative introduces a strict requirement. At the same time, the specific agency responsible for developing the format and procedure for applying warning labels is not yet named in the text—this leaves ample room for maneuver and potential disputes.
The key intrigue lies in the wording "created using." It could be interpreted broadly, covering not only fully generated materials but also those that have been edited by a neural network. The absence of a minimum threshold for the share of AI content in the draft only adds uncertainty. It is also unclear who bears responsibility for labeling—the creator or the distributor—and whether the label must "survive" the transfer of a file between platforms.
Arguments Against and Risks for Business
The initiative faces serious resistance in the expert community. The main thesis of critics: mandatory labeling will not become a panacea for deepfakes. Malicious actors using foreign services will easily circumvent the requirement, while Russian developers and users will gain an extra administrative burden. There is also a subtler problem: filters on photos are formally also the result of AI work, which, under a broad interpretation, could lead to absurd situations and worsen mass users' perception of the technology.
Particular concern is raised by the transfer of all significant decisions to subordinate regulations. According to a number of market participants, this creates a foundation for the unlimited expansion of the list of resources subject to regulation and blurs responsibility. Notably, large platforms with a daily audience of more than 500,000 Russian users are already obliged to technically ensure the ability to apply such labels. The framework article of the law is set to take effect on March 1, 2027, but the timeline for submitting the amendments themselves to the State Duma remains unclear.
International Context and Conclusions
Russia is not alone here. China is already moving along the path of a mixed model, where services add both visible and hidden labels, and platforms verify their presence. The European AI Act also obliges providers to ensure machine-readable labeling. However, the key difference is that foreign approaches are more detailed in terms of responsibility and technical implementation.
My expert view: the hasty introduction of mandatory labeling without clear technical standards and control mechanisms risks becoming a formality that creates the illusion of regulation but does not solve the problem of trust in content. The market needs not so much the norm itself, but a clear roadmap for its implementation, taking into account the interests of all participants—from developers to end consumers.