NewsGuard’s claim that AI chatbots like ChatGPT are amplifying Russian disinformation is facing serious pushback from researchers.
The issue started when NewsGuard published a March report alleging chatbots repeat false narratives from the Kremlin-linked Pravda network 33% of the time. The report tested major chatbots with prompts tied to Pravda’s stories, sparking headlines globally.
Soon after, critics called out NewsGuard’s opaque methods. The company refused to share its prompts, blocking independent review. Researchers also say the test was skewed—most prompts were designed to trigger falsehoods, inflating the 33% figure. 
Researchers ran their own tests on ChatGPT, Copilot, Gemini, and Grok. Their results? False claims showed up only 5% of the time. Just 8% referenced Pravda sites, and mostly to debunk them. Crucially, these references appeared mostly in queries with poor mainstream coverage.
The data points to “data voids”—rare topics lacking credible sources—rather than Kremlin efforts “grooming” AI. When chatbots lack solid info, they may pull from sketchy sites, but it’s not a Kremlin plot.
This means exposure to disinformation in chatbots requires a perfect storm: niche questions, few reliable sources, and chatbot fail-safes failing.
"Chatbots sometimes pull from dubious sites – not because they have been groomed, but because there is little else available."
The alarm around Kremlin “poisoning” AI could be overblown. Some experts warn Russian campaigns might fuel Western fears on purpose, overwhelming fact-checkers.
“Disinformation is a challenge – but so is the panic it provokes.”
This debate is critical as fears about AI and misinformation ramp up. Overhyping Kremlin influence risks misdirecting attention from real threats, like AI-generated malware flagged by Google and OpenAI.
The takeaway: Chatbots can repeat Kremlin talking points, but actual Kremlin manipulation of AI remains unproven and likely limited. The bigger problem may just be gaps in reliable data.
Read the full NewsGuard report here.
