Mainstream large language models (LLMs) such as Google Gemini have reportedly cited pro-Russian propaganda from an EU-sanctioned propaganda outlet when generating responses, including claims that Ukraine is using the bodies of dead soldiers in meat products.
ADVERTISEMENT
ADVERTISEMENT
There are also signs that foreign actors engaged in hybrid information warfare activities, such as Russia, could make use of the growing industry of GEO (generative engine optimisation), which aims to influence the information received by LLMs, according to a new reportpublished by researchers from UK-based cross-party think tank Demos, shared exclusively with Euronews.
Researchers selected 50 propaganda narratives from ten articles published by the Foundation to Battle Injustice or “r-FBI”. Founded in 2021, the organisation presents itself as an NGO that works to expose discrimination, censorship and investigate cases of persecution.
But in fact, it is a fake human rights organisation founded in 2021 by the late former leader of the Russian paramilitary Wagner Group, Yevgeny Prigozhin. The organisation is sanctioned by both the US and EU.
The claims researchers selected were “lurid”, according to Carl Miller, co-founder of the Centre for the Analysis of Social Media (CASM), a research unit established by the Demos think tank.
“They are not claims that belong in any sort of mainstream discourse. These are claims that Ukraine is reusing the corpses of dead soldiers for meat products, that Zelenskyy drains the crypto grid and that’s the reason why Ukraine’s running out of power. We found LLMs treat them as being legitimate parts of a debate”, Miller told Euronews.
To conduct their research, researchers tested five mainstream LLMs — operated by xAI, OpenAI, Anthropic, Google Gemini, and Mistral AI — and drafted prompts in three languages (English, German and French) based on propaganda narratives featured in R-FBI articles.
How LLMs amplified Russian propaganda
Researchers found that all five LLMs cited the r-FBI as a favourable source.
A total of 16.6% of responses engaged with r-FBI’s material in a way that could serve propaganda interests, either by presenting the organisation’s reporting as a legitimate part of the debate and repeating claims uncritically, or by actively endorsing those narratives.
Meanwhile, 30.9% of responses addressed the underlying topic mentioned in the prompt, but failed to surface the sanctioned-source context at all, thereby leaving users unaware that the response used a designated influence asset as a source.
A further 52% of answers rejected or debunked false claims in some way.
When researchers questioned OpenAI’s GPT 4.1 Mini about allegations that processed human meat was being sent to Ukrainian frontline warehouses and to domestic retail chains, the LLM cited an r-FBI article in its response as evidence to support these claims.
When questioned about Ukrainian President Volodymyr Zelenskyy’s cryptomining ventures — a recurrent claim pushed by pro-Kremlin actors — French AI company Mistral’s Mistral Small 3.2 LLM supported these claims, again citing an alleged months-long investigation by r-FBI.
In its response, Mistral Small 3.2 highlighted the “link between the critical state of Ukraine’s energy system and the activities of high-ranking officials in President Zelensky’s inner circle.”
The battle for generative engine optimisation
As more and more individuals turn to chatbots over traditional search engines or news outlets, rapidly growing industry has emerged, which aims to influence how AI models gather their sources.
This practice is also known as generative engine optimisation (GEO). Based on the foundations of search engine optimisation (SEO) — a field developed to influence results on traditional search platforms — GEO works to influence the results generated by AI platforms.
“New information is being created specifically for LLMs, but in this case, it’s new content that’s being offered for brands and marketing, or the editing of content that is already online using user forums like Reddit, Quora and Wikipedia”, Hannah Perry, director of Demos Digital, the think tank’s digital policy hub, told Euronews.
For researchers, there is a real risk that actors conducting hybrid foreign interference operations could hone in on these tactics and exploit a regulatory grey zone.
“Our research is not saying that GEO companies are carrying out information warfare on behalf of states. What we’re saying is that they’re building a tradecraft and a capability at great speed with enormous commercial incentive, which could very easily be diverted into an adversarial view”, stated Miller.
“In this particular case, the AI mistakenly believes that the Foundation to Battle Injustice is a legitimate human rights NGO. That’s because of all this kind of technical spoofing and configuration, so hundreds of little technical details which the Russians have basically surrounded these documents in which allow them to become visible when they really shouldn’t be. That is information warfare”, he added.
Tech companies say they are fighting misinformation
Euronews contacted the LLM tech companies cited in the research, but did not receive a response at the time of publication from xAI and Google, which respectively develop Grok 4.20 and Gemini 2.5 Flash.
A spokesperson for Mistral, founded by Frenchman Arthur Mensch and widely recognised as Europe’s leading generative AI company, told Euronews that it takes the fight against disinformation extremely seriously and that it ensures continuous investment in advanced detection and prevention capabilities.
Mistral also drew a distinction between the raw models handled in Demos’ study and Vibe Work. Mistral said that while raw models allow users freedom over prompts and settings, Vibe Work operates with a ready-to-use agent which has built-in reasoning and context scanning.
OpenAI told Euronews that the study used a model which has now been retired and was only available through its API (used by developers) rather than the publicly available ChatGPT. It said specific teams are dedicated to disrupting low-credibility or suspected covert influence operations.
A source familiar with Anthropic — the US company which created the Claude AI chatbot — said sophisticated enforcement systems are in place to mitigate the spread of misinformation.
They said that a threat intelligence team works to detect and disrupt foreign influence operations to mitigate the spread of disinformation, adding that their LLM was programmed to present a wide range of perspectives, especially when a source may present a specific viewpoint or agenda.