Moscow Turned AI into a Ghostwriting Machine for Global Influence
OpenAI banned a network of Russian-linked ChatGPT accounts used to covertly promote Kremlin talking points and fabricate an independent research community The operation centered on a fake Israel-based think tank called the International Policy Institute, with 34 of 36 articles being stolen from online sources AI was used to generate English and German-language content across X, LinkedIn, Facebook, and Telegram, including logos for ~12 Telegram channels with 10,000–20,000 subscribers each A key o
Analysis
TL;DR
- OpenAI banned a network of Russian-linked ChatGPT accounts used to covertly promote Kremlin talking points and fabricate an independent research community
- The operation centered on a fake Israel-based think tank called the International Policy Institute, with 34 of 36 articles being stolen from online sources
- AI was used to generate English and German-language content across X, LinkedIn, Facebook, and Telegram, including logos for ~12 Telegram channels with 10,000–20,000 subscribers each
- A key output was a self-styled "sovereignty index" designed to portray Russia favorably while disparaging nations opposing its invasion of Ukraine
- OpenAI assessed immediate reach as low but flagged the operation as part of a broader pattern of Russian AI-enabled influence infrastructure, including prior plans targeting African elections with up to $600,000 annually
Why It Matters
This case illustrates how state actors are operationalizing commercial AI tools for coordinated disinformation campaigns while actively evading detection through VPNs and origin concealment prompts. For AI practitioners and platform operators, it underscores the urgent need for robust detection of AI-generated influence operations that mimic legitimate academic or policy discourse. The incident also highlights the growing intersection of generative AI capabilities with geopolitical information warfare, making AI safety and integrity monitoring a critical infrastructure concern.
Technical Details
- Operators accessed ChatGPT through virtual private networks to bypass geographic restrictions and explicitly instructed the model to conceal Russian origin in generated content
- The fake International Policy Institute published 36 articles between September 2025 and May 2026, of which 34 were plagiarized from online sources, with the remaining two likely AI-generated
- Content was tailored across multiple platforms (X, LinkedIn, Facebook, Telegram) in both English and German, targeting audiences in Germany, the US, France, Poland, and Turkey
- AI was used to design logos for approximately a dozen Telegram channels, each accumulating between 10,000 and 20,000 subscribers
- The operation included a custom "sovereignty index" metric designed to rank countries based on alignment with Russian geopolitical interests, serving as a quantitative veneer for propaganda
Industry Insight
- AI companies must treat prompt-level instructions to conceal origin or impersonate entities as high-risk signals, warranting enhanced monitoring and account-level attribution rather than relying solely on content moderation
- The low immediate reach but strategic infrastructure-building nature of this campaign suggests that defensive focus should shift from measuring virality to detecting the establishment of persistent influence networks, including fake think tanks and indexed metrics
- The parallel report of fabricated video clips using AI-generated content and spoofed media outlets indicates a multi-modal escalation in Russian influence operations, making cross-platform and cross-modal detection capabilities essential for platform safety teams
Disclaimer: The above content is generated by AI and is for reference only.