NewsGuard audit: the 10 top AI chatbots, including ChatGPT, Claude, and Gemini, repeated false claims from a pro-Kremlin disinformation network 33% of the time
An audit found that the 10 leading generative AI tools advanced Moscow's disinformation goals by repeating false claims …
NewsGuard's Reality Check
Context & Ripple Effects
This is a recurrence rather than an isolated test: a 2024 examination of major chatbots similarly found Russian-disinformation repetition and citations to fake local-news sites, documented in an earlier cross-chatbot study.
The exposure sits within a widening synthetic-information problem. NewsGuard had previously identified AI-populated sites posing as news outlets, creating more low-quality material that can enter systems’ information environment.
First-order effects
The audit puts ChatGPT, Claude, Gemini and the other tested chatbots under immediate credibility pressure over responses to claims tied to the pro-Kremlin network.
Users seeking information on the tested claims may receive false narratives in a polished, conversational format, rather than a refusal or correction.
Second-order effects
Model providers face pressure to test for coordinated-disinformation sources and strengthen response safeguards, not only general misinformation policies.
The finding increases the value of independent evaluations such as NewsGuard’s, because provider assurances alone do not establish how models behave on adversarial prompts.
Third-order effects
If repeated audits continue to find the same failure mode, AI safety competition is likely to shift toward operational assurance: measurable, ongoing testing of model outputs and their information sources.
The combination of synthetic news-like content and conversational distribution could make provenance and source-quality controls a more central part of AI governance, though the audit alone does not show which interventions will work.
The trend: Generative-AI governance is moving from broad safety promises toward continuous, adversarial evaluation of how models handle coordinated information manipulation.
“NewsGuard's findings confirm a February 2025 report by the U.S. nonprofit the American Sunlight Project (ASP), which warned that the Pravda network was likely designed to manipulate AI models rather than to generate human traffic.” — www.newsguardrealitycheck.com/p/a- well-fun…
In incredibly unsurprising news: Pravda, a Russian propaganda network, has spent significant effort in LLM grooming - flooding the web with disinformation and aggregating it with the specific goal of influencing western AI outputs. — www.newsguardrealitycheck.com/p/a- well-fun.…
Claim of the moment: The top 10 leading AI chat bots advance Moscow's disinformation goals by repeating false claims from a pro-Kremlin propaganda network 33 percent of the time. https://www.newsguardrealitycheck.com/ ...
🇷🇺 A well-funded Moscow-based global ‘news’ network has infected Western artificial intelligence tools worldwide with Russian #propaganda — An audit found that the 10 leading generative AI tools advanced Moscow's #disinformation goals by repeating false claims from the pro-Krem…
@Techmeme This is the danger of closed source — These are knowledge models, and they only output what they are fed with — And no, they won't magically develop ‘reasoning skills’ and be able to sift through propaganda. NOT when it's part of the training — To think otherwise…
New report shows how ‘flooding’ the web with disinformation and propaganda affects AI models. — Deliberately infiltrating AI chatbots, by flooding search results and web crawlers with pro-Kremlin falsehoods, affects the AI models on current topics. — #propaganda #democracy #A…