VK’s AI profiles users as Russia expands digital surveillance

VK is pushing personalization beyond the familiar logic of “you liked this, so here is more of it.” The company’s emerging “neural profile” approach is designed to read sequences of behaviour across VKontakte, VK Video and VK Clips, treating a chain of views, comments and reactions as more meaningful than any single click.

That distinction matters. A person can watch one political video out of curiosity, open a fitness clip because a friend sent it, or like a post by accident. Repeated behaviour over time is different. It can show which subjects persist, which interests fade, how attention shifts and what kinds of content trigger stronger engagement.

VK has already built the infrastructure needed to make those signals portable across products. Its Discovery platform, launched in 2025, ties together recommendation and search systems. In July 2026, VK announced Discovery AI and said it could personalize results using interests drawn from different services in the group. In earlier corporate reporting, VK described a multimodal model that predicts user reactions and forms a viewer portrait from content interactions.

Materials accompanying the new rollout cite gains of 15% in content sharing, 10% in full-screen opens and 5% in likes. These remain company-attributed rollout metrics rather than independently verified performance data.

For a U.S. audience, the technology will look familiar. Meta, Google, TikTok and other large platforms have long used behavioural data to improve recommendations and ad targeting. The important difference is the surrounding political and legal system in which VK operates.

Russia’s information law requires a legal category of internet service providers known as organizers of information dissemination to retain specified user and communications data inside the country. As of Jan. 1, 2026, the retention period for certain metadata and user-action information covered by the rules was extended to three years. Message content can be retained for up to six months under government-set requirements.

Russia also operates SORM, the country’s technical interception system. In a 2026 regulation, the European Union described SORM as a surveillance infrastructure covering phone and internet communications, including social networks, with the Federal Security Service able to access copied traffic. The same document says the system can collect location, activity patterns and behavioural information. The Financial Times reported in June that the FSB’s role in controlling Russia’s internet had grown significantly since mid-2025.

None of that proves that the FSB receives VK’s finished internal “neural profile” as a ready-made dossier. VK’s public descriptions of Discovery do not say that, and the data-retention law does not identify a proprietary recommendation profile as a separate object that must be handed over. That distinction is important: a high-risk environment is not the same thing as evidence of a specific data pipeline.

The concern is more structural. When a private platform becomes better at turning everyday gestures into a stable model of a person, the value of the underlying data increases for anyone who can legally or technically access parts of that digital trail.

Research illustrates how much can be inferred. A 2024 paper in Scientific Reports used data from 1,358 consenting VK users to model personality traits and cognitive characteristics. The study had informed consent and was not a VK commercial system or a state surveillance project. But it demonstrates a broader principle: social-media behaviour can reveal characteristics that users never explicitly type into a profile field.

For VK, this is a product and advertising story. Better profiles can keep people engaged and make ad inventory more valuable. For Russian users, it is also a civil-liberties story because the data is created within an ecosystem of mandatory retention, centralized network controls and powerful security agencies.

The key issue is not whether one like is dangerous. It is what happens when thousands of small actions become a coherent behavioural history. That is the technical promise of VK’s new system — and the reason its political context matters as much as its recommendation accuracy.