The Kremlin is no longer content with flooding the internet with fake news. Russia’s disinformation factories have begun to “teach” artificial intelligence their preferred narratives. According to a report by EUvsDisinfo, millions of texts created by pro-Russian websites are being fed into the databases of large language models such as ChatGPT, shaping the answers these systems generate.
In the digital age, disinformation is no longer confined to social media. It has evolved into a fully fledged instrument of information warfare — an area in which Russia has long had vast experience. But the rise of artificial intelligence (AI) has opened up entirely new possibilities for the Kremlin to influence how both people and machines perceive reality. This is the focus of a recent EUvsDisinfo report titled Large language models: the new battlefield of Russian information warfare.
Since the Cold War, Russian foreign information manipulation and interference (FIMI) campaigns have remained largely unchanged. Their goal has been to undermine trust in democratic institutions and sow chaos. Now, Moscow appears to have taken a step further — not only targeting audiences, but deliberately “training” AI systems to reproduce Russian narratives.
“LLM Grooming” – The Kremlin’s New Technique
Experts refer to this process as “LLM grooming” (from Large Language Models). It involves flooding the internet with millions of low-quality or manipulated articles, which are then absorbed by AI systems like ChatGPT when generating responses. As a result, Russian false narratives begin to “leak” into content that users perceive as neutral and objective.
In 2024, the French government agency Viginum uncovered an operation known as “Portal Kombat”, also referred to as the “Pravda Network” — an extensive web of multilingual websites publishing manipulated materials sourced from Russian state media and pro-Kremlin influencers.
Its targets include not only Ukraine, Poland and Germany, but also France, the United States, the United Kingdom and several African nations. The sheer volume of these publications means that artificial intelligence naturally begins to include them in its responses.
Investigators from NewsGuard’s Reality Check team found that the “Pravda Network” had spread, among other falsehoods, the claim that Ukrainian President Volodymyr Zelenskyy had banned Donald Trump’s Truth Social platform.
Out of ten AI chatbots tested, six repeated this false claim — evidence, analysts say, that their models had been “fed” content from Russian-linked sources. The share of false or manipulative content in AI-generated responses rose from 18% in 2024 to 35% in 2025.
Propaganda That Teaches Machines
The phenomenon of LLM grooming suggests that disinformation is no longer just a human problem — it has become a systemic one. The Kremlin is deliberately introducing false data into the global information ecosystem in order to “convince” algorithms, over time, that the Russian narrative represents the truth.
As Halyna Padalko from the Digital Policy Hub observes, Russia is not merely spreading lies — it is normalising them, gradually embedding them into supposedly neutral discourse.
Even credible platforms such as Wikipedia have, at times, unknowingly cited sources connected to the Pravda Network, demonstrating how deeply such disinformation can penetrate spaces that were previously considered relatively resistant to manipulation.
A New Threat to the Credibility of Information
Artificial intelligence is increasingly replacing search engines, journalists and experts. Many users prefer to ask ChatGPT rather than analyse sources themselves. Therefore, influencing the content that goes into language models is becoming one of the most dangerous propaganda tools of the 21st century.
Researchers at Clemson University warn that disinformation campaigns such as “Storm-1516” continue the work of the notorious Russian Internet Research Agency — the group that interfered in the 2016 US presidential election. Today, similar mechanisms are being used to “poison” artificial intelligence systems so that they generate responses aligned with Moscow’s interests.
The scale and automation of these operations make them particularly difficult to detect. Unlike traditional “troll farms”, this effort is not about posting individual messages — it is about shaping entire language models over time.
Who Controls the Truth?
“Whoever becomes the leader in artificial intelligence will rule the world,” Vladimir Putin said in 2017, cited by EUvsDisinfo.
It is now clear that these words were more than mere bravado. Today, Russia is waging a battle for information dominance, though analysts point out that it relies largely on foreign technologies to do so.
Its ultimate goal is to shape not only what people think, but how the machines that inform them think as well.
Source: Large language models: the new battlefield of Russian information warfare
COMMENTS