A Moscow-based Russian disinformation network named 'Pravda' systematically infiltrated Western AI chatbots by flooding the internet with 3.6 million pro-Kremlin propaganda articles in 2024, causing major AI systems to repeat false Russian narratives 33% of the time.
A Russian disinformation operation called the Pravda network, administered by TigerWeb (a Crimea-based IT company owned by Yevgeny Shevchenko), launched in April 2022 following Russia's invasion of Ukraine. The network operates 150 websites across 49 countries in dozens of languages, publishing approximately 3.6 million articles in 2024 alone. Rather than targeting human readers directly, the operation uses search engine optimization to flood AI training data with pro-Kremlin falsehoods. NewsGuard tested 10 leading AI chatbots including OpenAI's ChatGPT-4o, Google's Gemini, Microsoft's Copilot, Meta AI, Anthropic's Claude, xAI's Grok, Mistral's le Chat, Inflection's Pi, You.com's Smart Assistant, and Perplexity. The audit found these chatbots repeated false Russian narratives 33.55% of the time, with seven chatbots directly citing Pravda articles as sources. The network has spread at least 207 provably false claims, including fabricated stories about Ukrainian President Zelensky and false claims about U.S. bioweapons labs in Ukraine. Despite receiving minimal human traffic (some sites have only 1,000 monthly visitors), the network successfully infiltrated Western AI systems through what researchers term 'LLM grooming' - deliberately manipulating AI training datasets.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed