AI content in Russia will begin to be labeled. Putin signed a law on regulating neural networks.

From September 1, 2026 (certain provisions — from March 1, 2027), large internet platforms with a daily audience of 500,000 or more people will be required to provide the technical capability for labeling audio, images, and video created using large foundational AI models. Labeling itself remains voluntary for users: the format and method of displaying the notice will be determined by agreements between AI service developers and their clients. However, the platforms' obligation to "provide the capability" is mandatory. This applies to social networks, major news sites, video services, and other high-traffic resources.

What exactly changes for the market

The law does not introduce blanket mandatory labeling of all AI-generated content. Instead, the state is creating an infrastructure of transparency. A user who generates material using a large model must have the technical capability to add a label. The platform is obliged to provide that capability. In practice, this means the emergence of new interface elements, API integrations with generative services, and, most likely, gradual pressure from regulators and audiences toward wider use of labels.

From an IT expert's perspective, this is a reasonable compromise. A complete ban or strict mandatory requirements at the current stage of technological development would create enormous technical and legal problems: how to prove content provenance, how to handle mixed (human + AI) material, and how not to crush smaller developers. Voluntary but technically supported labeling is a more flexible tool that leaves room for market self-regulation.

"National" and "sovereign" neural networks

The most important part of the law is the introduction of two statuses for large foundational models.

• Sovereign model — fully created by a Russian legal entity, controlled at all stages of its lifecycle, technically reproducible, and hosted in Russian data centers.
• National model — also developed by a Russian company, but may include open-source foreign components.

Both categories must comply with Russian legislation and "traditional spiritual and moral values." Developers of such models gain access to state support measures, including training data. In certain state information systems and sensitive sectors, the government will be able to impose requirements to use exclusively sovereign or national models.

This is a classic move toward technological sovereignty. Russia is not outright banning foreign models (as discussed in stricter early versions of the bill), but it is creating economic and regulatory incentives for developing domestic solutions. For businesses, this is a signal: in critical infrastructure, the public sector, and certain industries, the focus will be on local models. Companies already investing in YandexGPT, GigaChat, and similar projects get a clear growth trajectory.

What this means in practice

1. For platforms. Major players will need to refine their interfaces and, likely, establish content provenance verification processes. These are additional costs, but they are predictable and spread out over time.
2. For content creators. Labeling is still voluntary, but a culture of "honest AI" will take shape. In the long term, unlabeled AI content may be viewed with growing skepticism by audiences and advertisers.
3. For developers. A clear hierarchy of models is emerging. Those able to claim "national" or "sovereign" status will gain access to state support and government contracts. Others will have to compete in the open market.
4. For users. More transparency. Over time, it will become easier to distinguish generated material from "real" content. This is important amid the growing flood of deepfakes and synthetic news.

Expert perspective

The law turned out to be significantly softer than the Ministry of Digital Development's initial drafts. Strict bans on foreign models, mandatory labeling of everything, and a number of other controversial provisions were removed. Instead, the state has bet on supporting domestic development and building an infrastructure of transparency. This is a pragmatic approach.

In the global context, Russia is following a path similar to the European AI Act, but with an emphasis on sovereignty and localization rather than risk-based classification. The question is how effectively it will manage to combine support for domestic models with maintaining competitiveness. Large foundational models require enormous computing resources and high-quality data. State support will help, but it will not replace market competition and access to the best global practices.

In the coming months, we will see exactly how the government defines the criteria for "sensitive sectors" and support measures. Whether the new law becomes a real driver for the development of Russian AI or remains a framework document will largely depend on these subordinate regulations.

For now, the key takeaway is this: the era of unlabeled AI content in Russia's public space is gradually coming to an end. And that is, perhaps, the right step.