Russia's Project 2026 Targets AI and Search, Leaked Files Show

Leaked files obtained by Bloomberg reveal that Russia's Social Design Agency has been running a coordinated operation labeled "Project 2026" aimed specifically at seeding the information layer that search engines and AI chatbots draw from, rather than targeting news feeds or social timelines directly.

A planning document labeled "Projects 2026" describes several components. A German-focused Wikipedia clone is designed to look like legitimate reference material while embedding Russian propaganda, with the stated goal that AI platforms relying on publicly available text would absorb and repeat those narratives. A second component is described as an AI-driven "self-filling knowledge base" targeting Germany, for which the documents reportedly state that "servers have been launched and web shells created ... with the database already containing over 200,000 pages." A third initiative, apparently targeting Western think tanks, is described as already launched in English, with German, French, and Spanish versions planned.

The attack surface here is different from conventional bot campaigns. Social media posts can be flagged and removed; a reference platform that gets indexed and scraped into AI training datasets is far harder to unwind. According to the reporting, the intent is that propaganda could surface in AI-generated answers without an obvious link to its origin. Internal documents describe the broader program as carrying out "cognitive strikes" against Western societies, and the leaked materials indicate the Social Design Agency worked closely with officials inside the Russian Presidential Administration.

The caveats matter: these are leaked documents, and claims about what has been launched versus what remains planned should be taken as reported, not independently verified. What the reporting does not give you is confirmation from any major AI company that SDA-linked content has actually reached their training pipelines or live retrieval systems, nor a list of the specific platforms already indexed.

For practitioners building systems that retrieve from the open web, the operational implication is that source provenance and content authenticity are no longer background concerns. Retrieval-augmented systems that depend on web credibility have the clearest exposure, and any renewed push for AI training data auditing traces at least partly to state-directed operations like this one.