AI Content Labeling in Russia: Mandatory Standard May Become a Reality
Russian authorities are seriously considering a shift from voluntary to mandatory labeling of content created using large language models. This initiative, proposed by a working group under the Presidential Council for the Codification of Civil Legislation, could radically change the rules of the game for all digital market participants—from giants like Yandex to independent authors and streamers.
The new draft amendments to the AI law and Civil Code, already adopted in July, stipulate that images, audio, and video materials generated by neural networks must carry a mandatory label. Currently, this requirement is advisory in nature, but regulators appear determined. The key question is who exactly will be responsible for applying the label: the model developer, the distributing platform, or the end user. For now, the document's text is fraught with uncertainty, raising serious concerns among market participants.
What the amendments change
The phrase "created using" could be interpreted as broadly as possible. It may even cover materials that are merely edited by a neural network rather than generated from scratch. Moreover, no minimum share of AI-generated content in a work has been established. It is also not specified whether the label must be preserved when a file is transferred between platforms. Large platforms with a daily audience of more than 500,000 Russian users will be required to technically enable the application of such a label. The article of the framework law on labeling will take effect on March 1, 2027, but the exact timeline for submitting the amendments to the State Duma has not yet been determined.
Arguments against the mandatory requirement
The initiative already has influential critics. The main argument is that mandatory labeling will not help combat deepfakes. Malicious actors will continue to use foreign services and distribute materials without any labels. Moreover, the requirement could hurt legitimate Russian AI services by complicating their use without solving the problem. Technical feasibility also raises questions: how to preserve a machine-readable label amid numerous manipulations of content remains unclear.
International experience and developers' stance
It is telling that Russian developers are looking to international practice for guidance. The Chinese approach, where services add visible and hidden labels and platforms verify them, appears the most well-developed. The European AI Act also requires providers to implement machine-readable labeling and distributors of deepfakes to disclose the origin of content. Experts believe Russia may adopt a hybrid model. However, for now, the project remains at the discussion stage, and the final format of the requirement will depend on which option the amendment drafters and the authorized agency choose.
My view: mandatory labeling is an inevitable step in the era of generative AI, but haste here is dangerous. Without clear technical standards and explicit rules of responsibility, we risk getting an ineffective law that only creates additional barriers for business without protecting users from disinformation. The market needs a balanced approach that considers both developers' interests and society's need for transparency.