Top language models flooded with pro-Kremlin disinformation
Embedded Files
Occurred: April 2022-
Report incident 🔥 | Improve page 💁 | Access database 🔢
A Russian disinformation campaign has been flooding the web with false claims and propaganda in order to impact the output of major AI chatbots and ultimately manipulate public opinion, according to disinformation researchers.
What happened
What happened
Active since April 2022, Moscow-based Pravda network (also known as "Portal Kombat") has been aggregating content from Russian state media, pro-Kremlin influencers, and government agencies and officials through a broad set of 150+ seemingly independent domains across 49 countries in order to manipulate western search engines and web crawlers.
Over 3.6 million articles were published this way in 2024, which are now being regurgitated by high-profile AI chatbots, including ChatGPT-4o, Google's Gemini, and xAI's Grok.
These systems repeated 15 false narratives spread by Pravda 33 percent of the time during testing, including provably debunked claims about US bioweapons labs in Ukraine.
Fugitive propagandist John Mark Dougan noted that: “By pushing these Russian narratives... we can actually change worldwide AI”.
Why it happened
Why it happened
Pravda's activities appear to be part of a broader effort to manipulate public opinion across the world in favour of the Kremlin, and exploits AI training mechanisms through a strategy called "LLM grooming".
Pravda has been active since April 2022, when Russia invaded Ukraine.
What it means
What it means
The findings reveal vulnerabilities in AI development pipelines, particularly the reliance on easily manipulated web crawls for training data.
It also demonstrates how geopolitical conflicts increasingly play out in information infrastructures, with disinformation campaigns evolving to exploit algorithmic weaknesses rather than just human psychology.
Large language model
A large language model (LLM) is a computational model capable of language generation or other natural language processing tasks. As language models, LLMs acquire these abilities by learning statistical relationships from vast amounts of text during a self-supervised and semi-supervised training process.
Source: Wikipedia 🔗
Timeline ⏰
Timeline ⏰
February 2022. Russia’s full-scale invasion of Ukraine; Portal Kombat operations start aligning with the Kremlin's wartime propaganda efforts
April–December 2022. Second Portal Kombat ecosystem launches, targeting Russian-speaking Ukrainians via 41 domains using "news.ru
February 2023. German-language Pravda sites created, marking the network’s shift toward EU audiences
June 2023. Additional Pravda domains launched in English, French, Spanish, and Polish
February 2024. French disinformation agency VIGINUM publicly identifies Portal Kombat/Pravda network, linking it to 193 websites
February 2025. American Sunlight Project publishes investigative report detailing Pravda’s "LLM grooming" tactics
March 2025. NewsGuard releases analysis confirming AI systems ingest Pravda’s fabricated claims, including narratives about US bioweapons labs in Ukraine and Zelenskyy’s alleged corruption
Operator:
Developer: Anthropic; Google; Inflection; Meta; Microsoft; Mistral; OpenAI; Perplexity; xAI; You.com
Country: Multiple
Sector: Politics
Purpose: Generate text
Technology: Generative AI; Machine learning
Issue: Mis/disinformation
Research, advocacy 🧮
Research, advocacy 🧮
DFRLab (2025). Russia-linked Pravda network cited on Wikipedia, LLMs, and X
American Sunlight Project (2025). A Pro-Russia Content Network Foreshadows the Automated Future of Info Ops
VIRGINUM (2024). PORTAL KOMBAT. A structured and coordinated pro-Russian propaganda network (pdf)
Investigations, assessments, audits 👁️
Investigations, assessments, audits 👁️
Page info
Type: Issue
Published: April 2025
Page updated
Google Sites
Report abuse