Kremlin-linked Social Design Agency Projects 2026 seeds Wikipedia clones and fake think tanks to poison AI retrieval and search with Russian narratives ahead of European elections

Filed

September 11, 2026, 06:15 UTC

Actor

Social Design Agency (SDA), Moscow-based private agency sanctioned by US/UK/EU for Kremlin-directed information operations

Target

European and Armenian voters, search-engine and LLM users, Western think-tank and media audiences whose threat perceptions and electoral preferences on Russia, Ukraine, and regional alignments the seeded material is designed to shape

Source

Unidentified Source blurred: contractors, cutouts, buried disclosure

Confidence

high Multiple independent open sources, low disagreement about the facts

Techniques

Corpus poisoningNarrative launderingSynthetic mediaManufactured consensusFlooding the zone

Where

Europe, Armenia, France, Germany, Russia

Channels

• Wikipedia-style clone sites and reference pages (Armenia-focused examples: spyurk.cyou, sevan.info, khachkar.info)
• phony think-tank sites including World Center for Strategic Studies
• search-engine indexing and Common-Crawl pathways
• LLM training/retrieval corpora
• short-form video platforms (AI News initiative targeting France)
• opinion-leader social-media monitoring databases

Summary

Analysis

Primary open-source records are the 73 internal SDA documents (project proposals, chat screenshots, technical materials spanning May 2023–April 2026) obtained and reviewed by Bloomberg (23 June 2026), independently corroborated by European officials and researchers familiar with prior SDA leaks, and extended in the 10 September 2026 FDD/Cipher Brief analysis by Leah Siskind. Bloomberg verified 42 websites listed in the materials via DNS data; nearly all shared the same Russian IP. Armenia Wikipedia clones were created early 2026 and later blocked by hosting providers. A Germany component planned roughly 200,000 pages with monthly updates and explicit intent to ‘train’ six AI platforms. The World Center for Strategic Studies was registered March 2026 and began publishing unsigned analysis aimed at Western think-tank audiences, with language versions planned. Projects 2026 also called for a database of 10,000 European opinion leaders (100 top French opposition figures) and an ‘AI News’ stream of >500 short videos for French audiences.

Mechanism: corpus-poisoning is load-bearing. By seeding citation-shaped, encyclopedia-formatted text into the open web, the operation aims to make preferred frames appear as independent primary sources that search engines and LLMs will retrieve and amplify. Narrative-laundering occurs when the same material travels through performed think-tank and reference surfaces rather than labeled Russian state media. Synthetic-media and industrial page production supply volume no ordinary newsroom or research institute could sustain. Manufactured-consensus is attempted by density of converging ‘reports’ and reference entries. Flooding-the-zone saturates both human attention windows and machine-indexing cycles.

This is an influence operation and propaganda because origin is concealed behind institutional mimicry, the content is engineered for machine intermediaries the audience trusts more than RT or Sputnik, and the design goal is durable upstream distortion of foreign (primarily European) information environments rather than transparent advocacy. Technique scoring is actor-agnostic: identical Wikipedia-clone, synthetic-institute, and LLM-optimized seeding patterns appear in other state-linked campaigns already documented this cycle. Color is gray: the ultimate SDA/Kremlin linkage is visible in the leaked documents and sanctions records, yet the ordinary search or chatbot user encounters only the reference or think-tank surface.

Literacy counter: when a previously unknown reference site or ‘institute’ suddenly floods the indexed web with polished, byline-light, question- or encyclopedia-formatted material on contested geopolitical topics and begins appearing in search snippets or chatbot answers, inspect legal registration, staff pages, physical address, hosting/IP patterns, and any sanctions or prior leak linkages. Prefer primary government statements, named scholars with verifiable affiliations, and independent replication over volume-optimized synthetic dumps. Demand that provenance tags travel with every extracted paragraph. Apply the identical corpus-poisoning test to any faction’s LLM- and search-targeted reference networks—the method does not change with the flag.

Tells

• Sudden appearance of Wikipedia-style or think-tank sites with no staff, no address, and high page volume on contested topics
• Shared Russian hosting/IP patterns across ‘independent’ reference domains
• Question- or encyclopedia-formatted titles optimized for search and chatbot retrieval
• Explicit planning language around training AI platforms or influencing search rankings
• Rapid blocking of domains after journalistic exposure

Evidence

1. Bloomberg: Leaked Files Show Russia’s Plan to Influence AI and Search Results (23 June 2026)
2. FDD / Cipher Brief: Russia’s AI Blueprint Exposes an Authoritarian Playbook (10 September 2026)
3. RBC-Ukraine summary of Bloomberg SDA Project 2026 documents (23 June 2026)
4. OCCRP-linked reporting on Projects 2026 cognitive strikes and related operations