Officials have issued a warning regarding AI: groups linked to Russia would be trying to influence the responses given by artificial intelligence chatbots. This raises strong concern about the information that models may use in the future to generate responses.
As reported by the organization NewsGuard, pro-Russian websites spread versions according to which Armenian Prime Minister Nikol Pashinyan had allegedly tried to sell the gold from the Amulsar mine at a discount to Turkish companies. The issue is that the information turned out to be completely false; but what is worrying is that when AI chatbots were asked about the veracity of this information, they responded that it was true.
In the words of Chine Labbé, managing editor and senior vice president of partnerships for Europe and Canada at Newsguard: “In March 2025, we discovered that in 33% of cases, the leading commercial chatbots, including Mistral's chat and OpenAI's ChatGPT, repeated these narratives as verified facts, despite it being known that they are false stories serving the Kremlin's geopolitical interests.”
Almost a year later, in January 2026, Newsguard carried out a new investigation and, although some chatbots seemed to have improved, others continued to spread false information.
How does this strategy work?
The curious thing is that the method they would be using is not linked to hacking systems, but to the mass publication of propaganda content on the internet. As explained by the media outlet France 24, AI-powered chatbots are probabilistic tools, so they prioritize the most widespread information and not necessarily the most reliable.
In this way, AI models could take this information to train themselves and to generate responses to prompts. Specifically, European authorities point to “Pravda” as the pro-Russian propaganda network that would be behind this strategy. The network is made up of several websites that republish content aligned with the Kremlin. Previous investigations by NewsGuard — citing statistics from the non-profit organization American Sunlight Project — maintain that this network produced 3.6 million articles during 2024, with the aim of saturating the internet with disinformation and influencing the responses given by AI models. The organization examined 10 of the leading AI chatbots and found that they repeated false narratives of Russian disinformation.
A report cited by Wired had also warned about the same problem, stating that models such as ChatGPT, Gemini, DeepSeek, and Grok were reproducing Russian state propaganda when asked about the war in Ukraine. The outlet cited the Institute of Strategic Dialogue (ISD), which said that Russian propaganda took advantage of “data voids,” searches that yield few results, to spread misleading information.
For its part, NBC News noted that it consulted the companies developing the AI chatbots and some responded that they continually work to improve their systems, reduce hallucinations, and prevent them from reproducing false information.
The underlying concern is that the problem could worsen, given that more and more AI assistants consult information from the internet to formulate their responses. This is a concern that is not limited solely to Russia, but speaks to the new “fake news,” a strategy that any actor could use to manipulate the responses given by AI models.
