“A matter of details.” According to a report by the Institute for Strategic Dialogue (ISD), several large AI chatbots, ChatGPT (OpenAI), Gemini (Google), DeepSeek and Grok (xAI), unintentionally transmit elements of Russian state propaganda when asked about the war in Ukraine. The study reveals that around 20% of their responses cite pro-Kremlin or EU-sanctioned sources, such as RT, Sputnik or the Strategic Culture Foundation. Researchers believe these flaws may be explained by Russian propaganda exploiting gaps in data available online, particularly on sensitive topics where legitimate sources are scarce.
This situation worries the ISD, which highlights the risks linked to the growing use of chatbots as an alternative to search engines, particularly in Europe, where ChatGPT has more than 120 million monthly active users. The institute calls for better oversight of references to sanctioned media, while these models continue to use content from Russian disinformation networks. OpenAI responded by claiming to strengthen its detection and filtering mechanisms for manipulated sources to limit the spread of false information.
Interviewed by Wired magazine, Pablo Maristany de las Casas, ISD analyst and person responsible for the study, believes that “this raises questions about how chatbots should manage the referencing of these sources, knowing that many of them are sanctioned in the EU.”
He adds that “these results show the limits of large language models in restricting access to sanctioned media,” a growing problem as “more and more users turn to chatbots for real-time information.”
Several times affected by state propaganda.
Furthermore, this is not the first time that chatbots have been criticized for their reliability. In June 2024, a NewsGuard study revealed that the top ten chatbots, including ChatGPT, frequently transmit false information, particularly about the US elections, repeating elements of Russian propaganda spread on fake news sites.
These conclusions are added to other alerts, such as that of AlgorithmWatch in 2023, which denounced “immature and dangerous” technologies during the electoral period. For Reporters Without Borders, the observation remains the same: generative AI chatbots indiscriminately recycle trustworthy content and misinformation, blurring the line between truth and manipulation.
European authorities are also calling on artificial intelligence companies to better regulate their tools against the spread of Russian propaganda. A spokesperson for the European Commission recalled that “the affected providers must block access to the websites of the sanctioned media, including their subdomains,” while specifying that Brussels is working with national authorities to strengthen regulation.
“sounding boards for misinformation”
For its part, the Russian embassy in London rejected the report’s conclusions, stating that it was not aware of the cases cited. His spokesperson denounced “an attempt at political censorship” and an attack on freedom of expression, accusing Western governments of depriving the public of “independent” information, according to Wired.
For Lukasz Olejnik, a researcher at King’s College London, this controversy illustrates the limits of European regulation regarding artificial intelligence tools capable of aggregating online content in real time. “These systems become echo chambers, sometimes involuntary, for disinformation,” he explains, calling for closer cooperation between technological actors and public authorities.
Source: BFM TV
