- Filed
- September 8, 2026, 20:12 UTC
- Actor
- Russia-origin operators (OpenAI attribution; VPN access, Russian prompts)
- Target
- Western English- and German-speaking audiences, academic and policy discourse
- Confidence
- documented Primary documents, court records, official disclosures
- Techniques
- Synthetic mediaNarrative launderingTransferCard stacking
- Channels
- Chatgpt
- X (Twitter)
- Substack
- Telegram
- IBI website
Summary
OpenAI disrupted a cluster of ChatGPT accounts assessed as originating in Russia that generated English and German social posts promoting the International Burke Institute (IBI), a self-described Israel-based expert community. The IBI website, registered February 2025, hosted largely plagiarized academic articles with false attributions, listed real Western scholars who had never affiliated, and published a proprietary sovereignty index that ranked Russia favorably while criticizing Western governments and Ukraine policy. Reach remained modest (Breakout Scale level 3); infrastructure was elaborate.
Analysis
OpenAI’s 25 August 2026 disruption report supplies primary technical indicators: Russian-language prompting, VPN circumvention of geo-blocks, explicit instructions to the model to strip Slavic linguistic tells, and subsequent posting of the generated text by both branded IBI accounts and apparent everyday-user personas across five platforms. The non-AI core of the operation—the IBI website itself—exhibits classic narrative-laundering and transfer. Of a sampled 36 expert-linked articles published September 2025–May 2026, 34 were copied from legitimate academic or think-tank sources and re-attributed to incorrect authors, including real Western scholars (Fukuyama, Chomsky, and at least three CFR affiliates) who had never consented. A proprietary ‘Burke Sovereignty Index’ and accompanying country reports systematically scored Russia high and Western governments low, often in polemical language that framed EU membership or support for Ukraine as loss of autonomy. This is card-stacking by selective metrics and framing rather than balanced comparison.
The combination qualifies as influence activity, not mere opinion. The deception is structural: false institutional identity (Israel-based expert community), borrowed authority of named scholars, and AI-assisted persona and content generation intended to insert pro-Moscow frames into Western information streams under the appearance of independent analysis. Low audience reach does not negate the technique set; the same infrastructure could be scaled. Linguistic artifacts (e.g., ‘Svetofor coalition’ as literal translation of the German Ampelkoalition) and the mismatch between claimed authors and actual expertise provide forensic tells independent of platform attribution.
Scoring methods rather than nationality keeps the analysis neutral: identical techniques—synthetic personas, reputation transfer, selective framing, and multi-platform laundering—appear across state and non-state actors. The episode also illustrates the value of AI-lab threat reporting, which can surface activity before it matures into high-volume amplification.
Literacy counter: treat newly appeared ‘expert institutes’ that list prestigious names without verifiable institutional footprints, publish large volumes of recycled content under mismatched bylines, or promote proprietary indexes that systematically flatter one geopolitical camp with heightened skepticism. Cross-check claimed affiliations directly with the named scholars or their home institutions. Prefer primary documents and platform technical reports over secondary political framing. When AI-generated text is suspected, look for residual translation artifacts, unnatural register shifts, and coordinated multi-platform posting patterns that share infrastructure or timing.