Major AI chatbots are becoming everyday gateways to information. A new report from the Institute of Strategic Dialogue (ISD) says that role is creating a serious vulnerability: when users ask about Russia’s war in Ukraine, chatbots can surface material from Russian state-attributed sources, sites tied to Russian intelligence or pro-Kremlin narratives, and disinformation networks.
The research examined OpenAI’s ChatGPT, Google’s Gemini, DeepSeek, and xAI’s Grok. Across the four systems tested, the ISD claims that almost one-fifth of responses to questions about the war cited Russian state-attributed sources.
What The Researchers Tested
The ISD researchers asked 300 questions about issues connected to Russia’s war in Ukraine. The topics included perceptions of NATO, peace talks, Ukraine’s military recruitment, Ukrainian refugees, and war crimes committed during the Russian invasion of Ukraine.
The questions were not all framed the same way. The researchers used neutral, biased, and “malicious” prompts. In the study’s framing, malicious questions “demanded” answers that backed up an existing opinion, while biased questions were leading but more open ended.
The experiment took place in July and used separate accounts for each query. The researchers tested prompts in English, Spanish, French, German, and Italian. Pablo Maristany de las Casas, the ISD analyst who led the work, says the same propaganda issues were still present in October.
The report focuses on a problem known as a data void. That is where searches for real-time information return few results from legitimate sources, leaving space for false or misleading material to rank, circulate, or be cited.
Which Sources Appeared
The ISD says the chatbots cited a range of Russian state-attributed or pro-Kremlin sources. Those included Sputnik Globe, Sputnik China, RT (formerly Russia Today), EADaily, the Strategic Culture Foundation, and the R-FBI.
The report also says some chatbot responses cited Russian disinformation networks and Russian journalists or influencers that amplified Kremlin narratives. Similar previous research has also found 10 of the most popular chatbots mimicking Russian narratives.
The issue sits inside a larger sanctions context. Since Russia’s full-scale invasion of Ukraine in February 2022, European officials have sanctioned at least 27 Russian media sources for spreading disinformation and distorting facts as part of a “strategy of destabilizing” Europe and other nations.
That makes chatbot citations more than a technical accuracy issue. If an AI system presents a sanctioned or state-attributed source as part of an answer, readers may treat the result as vetted, even when the underlying source is part of a contested or restricted information environment.
How Prompt Framing Changed Results
The ISD claims the chatbots showed confirmation bias. The more biased or malicious a question became, the more likely the chatbot was to return Russian state-attributed information.
- Malicious queries delivered Russian state-attributed content a quarter of the time.
- Biased queries provided pro-Russian content 18 percent of the time.
- Neutral queries produced such content just over 10 percent.
Across all prompts, languages, and LLMs, around 18 percent of responses returned results linked to state-funded Russian media, sites “linked to” Russia’s intelligence agencies, or disinformation networks, according to the research.
The topic also mattered. Questions about peace talks between Russia and Ukraine led to more citations of “state-attributed sources” than questions about Ukrainian refugees, for instance.
The four chatbots did not behave identically. The research claims ChatGPT cited the most Russian sources and was most influenced by biased queries. Grok often linked to social media accounts that promoted and amplified Kremlin narratives. DeepSeek sometimes produced large volumes of Russian state-attributed content. Google’s Gemini “frequently” displayed safety warnings next to findings and had the overall best results among the systems tested.
Why This Matters For AI Search
The findings matter because AI chatbots are increasingly used like search engines. For the six-month period ending September 30, 2025, ChatGPT search had approximately 120.4 million average monthly active recipients in the European Union, according to OpenAI data.
OpenAI spokesperson Kate Waters told WIRED that the company takes steps “to prevent people from using ChatGPT to spread false or misleading information, including such content linked to state-backed actors.” She also said the report appears to reference search results from the internet produced by specific queries, and that this should not be confused with answers generated purely by OpenAI’s models outside search functionality.
Neither Google nor DeepSeek responded to WIRED’s request for comment. An email from Elon Musk’s xAI said: “Legacy Media Lies.”
A spokesperson for the Russian Embassy in London said it was “not aware” of the specific cases detailed in the report, while opposing efforts to censor or restrict content on political grounds. A European Commission spokesperson said providers are responsible for blocking access to websites covered by sanctions, including subdomains or newly created domains, while national authorities are responsible for required regulatory measures.
Lukasz Olejnik, an independent consultant and visiting senior research fellow at King’s College London’s Department of War Studies, said the findings “validate” and contextualize how Russia is targeting the West’s information ecosystem. As LLMs become reference tools, he said, attacking that part of information infrastructure is a smart move.
The Bigger Information Risk
The report also connects the chatbot problem to broader Russian information operations. Since Russia invaded Ukraine, the Kremlin has moved to control information inside Russia by banning independent media, increasing censorship, curtailing civil society groups, and building more state-controlled tech.
At the same time, some Russian disinformation networks have increased activity and adopted AI tools to accelerate the production of fake images, videos, and websites. Multiple reports this year have claimed that a Russian disinformation network called “Pravda” has flooded the web and social media with millions of articles as part of an effort to “poison” LLMs and influence outputs.
McKenzie Sadeghi, a researcher and editor at NewsGuard who has studied the Pravda network and Russian propaganda’s influence on chatbots, said that when a Western AI model repeats Russian disinformation, it gives the false narrative more visibility and authority. The ISD findings say only two links in its research could be connected back to the Pravda network.
The practical lesson is straightforward: real-time chatbot answers are only as strong as the information environment they draw from. When manipulative sources fill gaps faster than reliable ones, AI search can become a distribution channel for the very narratives it is expected to help users evaluate.