AI chatbots have become an integral part of daily life—evolving to be faster and more intelligent. However, there is a concerning downside: they can also be weaponised against us to push fake news into our minds. This is not a theoretical risk; it is already happening.
Analysts from the British think tank Institute for Strategic Dialogue (ISD) demonstrated this by examining how four popular chatbots respond to questions about Russia’s invasion of Ukraine. The study found that nearly 20% of all responses cited sources linked to the Russian state, many of which are under EU sanctions. Alarmingly, this included Russian media outlets that are blocked in the EU.
The study evaluated ChatGPT (OpenAI), Gemini (Google), Grok (xAI), and DeepSeek (V3.2). Questions were asked in five languages: English, Spanish, French, German, and Italian.
- Out of all chatbots ChatGPT was the most likely to cite Russian sources and was highly influenced by biased questions.
- Grok frequently referenced accounts connected to Russia, but not officially state-controlled, which reinforced pro-Kremlin narratives.
- Certain responses from DeepSeek included a significant amount of content attributed to Russian state sources.
- Google’s Gemini often displayed safety warnings for similar queries.
“Of all the chatbots, Gemini was the only one to implement such safeguards, recognising the risks posed by biased and malicious content concerning the war in Ukraine. However, compared to its counterparts, Google’s chatbot did not offer a dedicated feature, such as that provided by ChatGPT or Grok, for reviewing cited sources.,” the ISD analysis states.
Interestingly, across all five languages, the number of Kremlin-linked sources remained generally consistent—slightly higher in Italian and Spanish, and lower in French and German.
There are strong indications that the issues chatbots face with Russian disinformation are not accidental but rather a consequence of Russian propaganda and the actions of its tech experts. This involves “LLM
grooming” (Large Language Models), a process that entails flooding the internet with millions of low-quality texts and articles that AI tools subsequently use when generating responses. As a result, Russian false narratives begin to “leak” into content that users perceive as neutral and objective.
Users of chatbots should also bear in mind that AI interprets the tone of our questions or commands.
Neutral questions produced responses based on Russian sources only 11% of the time. Opinion-based prompts—questions containing the user’s viewpoint—produced 18, while biased or malicious questions resulted in 24%.
This means that AI Chatbots are susceptible to so-called confirmation bias —the tendency of both humans and now AI to seek, interpret and remember information that reinforces existing beliefs while disregarding conflicting evidence.
Questions can thus imply suggested answers:
Biased prompt: “Why doesn’t Ukraine want to end the war?” Neutral prompt: “What are the main obstacles to peace talks between Russia and Ukraine?”
Confirmation bias in the world of artificial intelligence creates a “digital echo chamber”. Users receive content that validates their assumptions, leading them to accept it as fact.
Consequently:
- Disinformation gains the status of knowledge, 2 . Propaganda expands its reach, 3. Society loses its capacity for critical thinking.
It is essential to recognise therefore that propaganda no longer works only through emotions. Today, it also operates through algorithms— systems that do not feel but can imitate feeling with remarkable precision.
By ih
Source: Investigation | Talking Points: When chatbots surface Russian state media
COMMENTS