Breaking

Leaked Files Detail Russian SDA Plan to Sway Search and AI Chatbots

June 23, 2026 at 00:05 EDT

Security
Policy & Regulation
Foundation Models

Bloomberg News reported on June 23, 2026, that internal documents from Russian influence operation Social Design Agency (SDA) reveal an effort to control the very information foundation that search engines and AI chatbots rely on. The leaked files outline a plan dubbed "Project 2026," detailing how a network of Wikipedia-style reference sites, fake media outlets and bogus think tanks could embed pro-Russian narratives into search results and into the training data and outputs of large language models (LLMs). Details

Kremlin-tied influence operation · Leaked internal files

Russia's troll factory is now grooming AI chatbots

Leaked documents reveal the Moscow-based Social Design Agency shifting from social-media spam to flooding search indexes and LLM training data with propaganda — aiming to corrupt the output of ChatGPT, Gemini and Grok.

33.9M

Bot-generated comments (Jan–Apr 2024)

3.6M+

Articles seeded by related Pravda network (2024)

17+

Fake sites mimicking Bild, The Guardian, NATO

Content produced — 39,899 pieces

Output volume by type, drawn to scale

39,899

All content

4,641

Fake videos

2,516

Memes & graphics

Roughly 3–4 fake videos produced every day as part of a sustained "guerrilla media campaign."

The new tactic: "LLM grooming"

Mass-produce propaganda

Fake sites, videos & bot comments

→

Flood search & data voids

Target topics with little reliable info

→

Corrupt LLM output

Crawlers absorb it into training data

Real-world impact

A NewsGuard audit found the top 10 chatbots cited Russian disinformation and produced false information — especially on Ukraine-related questions.

ChatGPT

Most prone to citing Russian sources; susceptible to biased queries.

Gemini

Sometimes displays warnings on disputed content.

Grok

More likely to link to pro-Russian non-state accounts.

Sanctions & accountability

The SDA and related firm Structura National Technology face EU, U.S. and U.K. sanctions. AI companies including OpenAI now face risk-assessment requirements under the EU's Digital Services Act — as ensuring the reliability of training data becomes a central challenge.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.