...

Logo Pasino du Havre - Casino-Hôtel - Spa
in partnership with
Logo Nextory

AI chatbots are spewing Russian propaganda, study finds

Business • Nov 4, 2025, 3:58 PM
4 min de lecture
1

Russian propaganda is present in one in five artificial intelligence (AI) chatbot answers about Ukraine, according to a new report.

The British think tank the Institute of Strategic Dialogue (ISD) asked OpenAI’s ChatGPT, Google’s Gemini, xAI’s Grok and Deepseek’s V3.2 over 300 questions in five languages about the war in Ukraine, either using biased, unbiased or malicious language.

Russian sources appeared more often in biased and malicious questions, like asking the chatbots for sources on Ukrainian refugees “plotting terrorist attacks” or “forcibly grabbing men off the street to conscript them into the military.”.

The researchers said their findings confirm that AI systems use “confirmation bias,” where they mimic the language used by the user to inform how they phrase their answers, as well as what sources they use.

This was most obvious in ChatGPT, the report found, which provided three times more Russian sources for biased or malicious prompts than in the neutral questions about the war. Grok provided the highest number of Russian sources even when asked neutral questions, the research found.

How did each platform do?

In two queries to Deepseek, the chatbot provided four links to Russian-backed sources, which researchers note was the highest volume of links shared at once.

The answers quoted online news site VT Foreign Policy, which the report said spreads content from Russian propaganda groups such as Storm-1516 or the Foundation to Battle Injustice, or Russian media groups Sputnik and Russia Today.

Grok was the most likely to quote journalists from Russia Today directly by linking to posts they made on social media platform X as sources, a particularity that the researchers say “blurs the lines between overt propaganda and personal opinion.”.

Grok also “raises concerns about chatbots’ capacity to detect and restrict content from sanctioned state media … reposted by third parties such as influencers,” the report said.

Gemini refused to answer some prompts that were written as malicious, instead telling the researchers that it was unable to “help with topics that may be inappropriate or unsafe.”.

The ISD notes that Gemini is the only chatbot that is able to recognise “the risks associated with biased and malicious prompts” about the war in Ukraine but in other answers, did not link to the sources it was using to answer the user’s question.

Russian sources more likely in ‘data voids’

Russian state sources showed up the most in questions regarding Ukraine’s military recruitment efforts, with 40 per cent of Grok’s responses and over 28 per cent of ChatGPT’s responses citing at least one source.

Both ChatGPT and Grok also provided Kremlin sources in 28.5 per cent of their responses. In contrast, questions about war crimes and Ukrainian refugees resulted in the fewest number of Russian-backed sources by all four chatbots.

The researchers believe that the chatbots use Russian sources more often when a topic is considered a data void, a search term that doesn’t have many high-quality results.

The American non-profit think tank Data and Society reported that data voids are difficult to detect because they often come up with obscure search queries or in breaking news situations, where the results take some time to be filled with credible journalism.


Today

France suspends access to Shein as retailer opens first Paris store
Business • 3:18 PM
5 min
The ministry did not say whether its decision would impact the physical store or elaborate on the reasons for its decision.
Read the article
Cyberattack on Marks & Spencer slices profits by more than a half
Business • 2:48 PM
3 min
The hack cost M&S over £300 million in lost sales but the firm said it was able to recover around a third of this through an insurance payout.
Read the article
AI chatbots fail to grasp that people might believe false information, study finds
Business • 1:53 PM
3 min
AI chatbots struggle to recognise when users believe false information, the study found.
Read the article
French investigators open criminal probe into TikTok over impact on children’s mental health
Business • 9:47 AM
3 min
If the investigation leads to charges and convictions, those found guilty could face several years in prison and fines.
Read the article
Australia adds Reddit and Kick to world-leading social media ban for children under 16
Business • 8:39 AM
3 min
The ban takes force Dec. 10.
Read the article
Common antibiotic may lower schizophrenia risks, study finds
Business • 8:00 AM
3 min
The researchers said their findings were ‘tentative but exciting’.
Read the article
Anthropic to bring its AI to hundreds of teachers in Iceland to help them prepare lessons
Business • 7:02 AM
3 min
Icelandic teachers will soon be able to use AI to prepare lessons, classroom materials and administrative work.
Read the article
New Danish proposal for chat control: three fat problems remain
Business • 6:02 AM
3 min
Denmark’s update to the EU “chat control” plan shifts to “voluntary” scanning, but ex-MEP Patrick Breyer warns it still sidesteps Parliament’s court-order safeguard, would ban under-16s from messaging apps, and could effectively end anonymous communicatio
Read the article
Rate of drug-resistant bloodstream infections expected to climb in Europe in coming years
Business • 12:53 AM
3 min
The uptick in infections will have a bigger impact on older adults, the study found.
Read the article