Russian lawmakers are seriously considering a transition from voluntary to mandatory labeling of content created using large language models. This concerns amendments to the recently adopted AI law and the Civil Code, which are currently being actively developed in working groups.
According to the prepared draft, users and companies will be required to label images, audio, and video generated by neural networks. The initiative comes from the working group of the Presidential Council on the Codification of Civil Legislation, led by Pavel Krasheninnikov, Chairman of the State Duma Committee on State Construction. Consideration of the amendments is scheduled for August 19.
Key changes and uncertainties
The current version of the AI law does not contain mandatory labeling requirements, but the new draft radically changes the approach. The format, content, and placement procedure for the warning will be determined by an authorized state body, although the specific agency has not yet been named. An important nuance: the wording "created using" could also cover materials merely edited by a neural network, while the minimum share of AI content in a project has not been established.
The question of responsibility remains open: who should place the label—the author or the distributor? The draft does not introduce special liability and does not clarify who will verify the presence of the label. It is also unclear whether the label will be preserved when a file is transferred between platforms. In the current version, creators of audio and visual content using foundation models are only required to provide the ability to place a warning.
Large platforms with a daily audience of more than 500,000 Russian users are obliged to provide the technical capability for such labeling. The labeling requirement is set to take effect on March 1, 2027, but the timeline for submitting the amendments to the State Duma has not yet been determined.
Arguments against and international context
Participants in the discussion are skeptical of the initiative. In their view, mandatory labeling will not become a panacea for deepfakes: malicious actors will continue to use foreign services and distribute unlabeled materials. Moreover, the measure could complicate work with Russian AI services without solving the root problem.
Experts also point to the prematurity of such requirements: the world is still studying how technically feasible it is to preserve a machine-readable label during further manipulation of content. Significant decisions are being delegated to subordinate regulations, which creates additional legal uncertainty. Representatives of technology companies fear that constant reminders about labeling will only worsen users' perception of AI technologies.
Meanwhile, international experience shows: China has already implemented a mixed model with visible and hidden labels, while the European AI Act requires providers to use machine-readable labeling and disclose the origin of deepfakes. Russia should likely move in the same direction, but taking local specifics into account.
My view: The initiative is logical, but its implementation raises questions. Without clear definitions of responsibility and technical standards, mandatory labeling risks becoming a formality that is easy to circumvent. The market needs not so much new prohibitions as working content verification mechanisms that do not create an excessive burden on businesses.