FSB Commissioned AI to Track Opposition, 'Spider' Project Collapsed After 18 Months

For a year and a half, Russia's Federal Security Service (FSB) attempted to build an artificial intelligence system designed to automatically track down Kremlin critics and protest participants. The project, codenamed PAUK (Spider), was commissioned by the FSB's Second Service, known as Dvoyka, a unit previously linked to the poisoning of Alexei Navalny. The system never became operational in its intended form, and its details only came to light after hackers breached the contractor.

What Spider Was Meant to Do

According to documents described by iStories, PAUK was supposed to scan Russian and foreign media, anonymous Telegram channels and social networks every day, then automatically prepare concise security threat summaries for FSB leadership. The key feature was meant to be cross-referencing open-source information with the service's classified databases.

Eventually, officers were to be able to query the system like a chatbot about specific individuals deemed "of interest" and receive in response a full name, date of birth, phone number or car registration number. Such a tool was meant to drastically speed up work that previously required analysts to manually search registries.

The system mainly monitored activity related to protests against utility fee hikes, opposition to environmentally harmful investments, and the activities of migrants in Russia. Among the tracked sources, independent media outlets, including the Activatica website, accounted for as much as one-sixth of all entries in the summaries the system generated in the first half of 2025.

The Unit Behind Navalny's Poisoning

The FSB's Second Service, for which the contractor formally worked, is the same unit that investigations by Bellingcat and partner outlets had previously linked to the 2020 poisoning of Alexei Navalny with the Novichok nerve agent. Commissioning an AI system to automatically profile activists thus fits into a broader, already-known pattern of behavior for this unit, this time carried out with the help of a commercially available language model.

The formal client was the FSB's Center for Information Security and Special Communications, but the system's actual end user was meant to be Dvoyka. This kind of division is used within Russian security structures to separate operational units from contracts with outside companies, which later makes it harder to prove direct responsibility.

Why the Project Collapsed

Despite its ambitions, the project was never deployed in its planned form. According to an account cited by iStories, the team at Kazan-based Mikord lacked sufficient experience working with large language models, and the company missed its deadline for delivering a functioning system. After eighteen months of work, in autumn 2025, the project was suspended.

Farewell, project - from Mikord's internal correspondence, revealed by iStories

The Spider story shows that access to a ready-made open-source model like Llama 2 is not enough on its own to build a functioning surveillance system. Deployment requires a team that understands the limitations of such models, can fine-tune them for specific tasks, and can verify the quality of the generated responses, something that was missing in this case.

How the Story Came to Light

The project was exposed by chance. In December 2025, an anonymous hacker group breached Mikord's systems and stole internal documents, which were then passed to organizations helping Russians avoid military conscription. The materials eventually reached journalists at iStories, an independent investigative outlet operating outside Russia, who reconstructed the project's history and contents.

The case was later also covered by Meduza's editorial team, who pointed directly to the client's connection to the unit responsible for earlier persecution of Kremlin critics. For the West and for Poland, it is further evidence that Russian security services treat artificial intelligence as a tool for mass surveillance of their own citizens, not only for operations aimed abroad, such as disinformation or infrastructure attacks.

The failed, for now, Spider project does not mean the FSB will abandon further attempts to automate surveillance. Russian security services regularly test new tools based on open-source data, and the leaked documents reveal the scale of ambition: a thousand monitored sources and a database of queries covering activists' personal data. Future versions of similar systems may no longer have the same weaknesses as this first attempt.