About one in every three responses from major AI chatbots contains false narratives connected to a Russian disinformation network, a recent audit shows. The alarming figure comes from a March 2025 investigation by NewsGuard, a firm specializing in media credibility. It reveals how chatbots inadvertently echo Kremlin-backed propaganda not due to malicious intent but because they absorbed polluted data from the web during training.
Pravda Network's Massive Influence on AI Responses
The disinformation originates from the Pravda network, an extensive operation spanning around 150 websites, publishing misleading content in 46 languages. Analysts estimate the network produces some 18,000 articles per false claim. When AI models scan billions of online sources, the sheer volume of repeated misinformation gives these false claims undue weight, skewing the bots' outputs.
The companies whose AI models showed this vulnerability include tech giants like OpenAI, Google, Microsoft, and Meta. In several cases, chatbots even directly cited Pravda-linked websites as their sources.
Heightened Risk Around Elections and Northern Europe
Additional data from a France 24 report highlights an even more concerning trend: in responses to queries in Nordic languages, disinformation prevalence jumped to nearly 50%. This surge has triggered formal warnings from European authorities, citing risks that pro-Russian actors are deliberately seeding AI training data ahead of key electoral events in the region.
The looming EU AI Act aims to tackle these issues by demanding greater transparency about AI training data sources and introducing stricter auditing standards. However, experts warn that thoroughly vetting the massive datasets feeding these AI systems will be a complex and costly task, posing challenges for compliance and content integrity going forward.
