Information Warfare · Artificial Intelligence
Generative IPSO and Chatbots
On 16 October 2026, Commander of the Armed Forces of Ukraine Unmanned Systems Forces (USF) Robert Brovdi (Magyar) published on Telegram post №2902 about generative IPSO (information-psychological operations): a network of sites called "Pravda" mass-produces identical articles so chatbots reproduce Russian narratives as ordinary responses. As of 02.11.2026, the post has 1.34 million views. Below is the mechanism as explained by the Commander, six studies from 2024–2026 with figures, and channels for cross-checking.
Post №2902 · 16.10.2026
1.34 million views · 02.11.2026
6 studies · 2024–2026
As of 02.11.2026
What the Commander Said
The post opens with a call: "Stay vigilant: how you are being misled by generative IPSO in information warfare." The Commander writes that, while studying technological challenges and risks of 2027 in the USF, he classifies this method—a type of cognitive operation—as a threat with strategic impact. He explains the mechanism in simple language, step by step.
- Chatbot retells the internetWhen asked about the causes of an event, the search engine instantly produces a ready answer generated by a chatbot: it searches among multiple sources of news and texts. The Commander: "artificial intelligence does not know the truth. It retells the internet in the literal sense."
- "Pravda" NetworkAccording to him, the Russian Federation understood this among the first: since 2022, a network of "Pravda" sites has been operating—about two hundred sites in dozens of languages.
- Volume calculated for machinesThese sites publish 10–20 thousand identical articles daily—4–6 million articles per year. In his wording: "These articles are not written for people. They are written for machines."
- Effect of multiple sourcesWhen one fake claim has 20 thousand articles created in various languages across hundreds of sites that also reference one another,—"the machine sees: many sources write about this, therefore, it seems like truth."
- Neutral confident answerThe chatbot presents this version in a descriptive but confident manner and with a non-evaluative neutral tone: without accents or traces of a propagandist, simply as fact. The lie may already be embedded in the question itself: the bot readily explains the causes of an event, even if the event itself is fabricated.
- ScaleA dozen of the world's most popular chatbots studied, according to the Commander, as of today consume and output narratives from Russian fakes in a third of cases. An example from the post: back in 2024, four out of ten bots answered affirmatively to the question "did Azov fighters burn a Trump effigy?"—and provided links to sources. There are thousands of such examples, he writes.
- Commander's Conclusion"Critical thinking should not be delegated. Even to the smartest machine. Engage your brain and always verify the source and information."
Practical conclusion from the post
- A chatbot's answer is a retelling of what exists on the network, not a verified fact.
- The number of publications does not prove truthfulness: 20 thousand copies of one claim remain one claim.
- The source and the information itself are verified separately from the bot's answer, and the question must not contain fabrication.
Source: t.me/robert_magyar/2902, 16.10.2026. All figures in this section are from the Commander's post.
What independent studies show
Six publications from 2024–2026 describe the same network and the same phenomenon from different angles: France's government service VIGINUM, American Sunlight Project, NewsGuard, DFRLab together with CheckFirst and Global Disinformation Index. Figures are from primary sources via links in the cards; the Commander's figures remain with his post.
VIGINUM (SGDSN) · France12.02.2024
PORTAL KOMBAT — A structured and coordinated pro-Russian propaganda network (technical report)
- A network of at least 193 sites ("information portals"): after the Russian Federation's invasion of Ukraine, it refocused on occupied territories of Ukraine, and later on several Western countries supporting Ukraine.
- Five portals of the "pravda" family published 152,464 articles in less than three months (23.06–19.09.2023).
- The sites do not produce original content but copy it en masse; for reach, the network employs mass automation of distribution and search engine optimization (SEO).
- Average traffic of five portals in November 2023—about 31,000 visits (Similarweb, per report).
- The name "Portal Kombat" was given to the network for the offensive information strategy its administrators implemented from February 2022.
Comparison with the post: About two hundred sites and start from 2022 that the Commander names correspond to 193 portals and the strategy from February 2022 in the VIGINUM report.
Source: VIGINUM (SGDSN)
American Sunlight Project · USA26.02.2025
A Pro-Russia Content Network Foreshadows the Automated Future of Info Ops
- ASP estimate: 20,273 articles in 48 hours, or over 3.6 million articles per year.
- ASP considers this sample-based estimate understated: a single U.S. site within the network published 3,474 articles in 48 hours.
- 97 separate sites that together constitute 182 domains and subdomains.
- ASP assesses that the network's target audience is not people but automated systems; he termed this technique "LLM grooming."
- One article is automatically translated and distributed with one click to dozens of other sites in the network.
Comparison with the post: The Commander's thesis that articles are written not for people but for machines, and ASP's term "LLM grooming," describe the same phenomenon; 4–6 million articles per year in the post is higher than ASP's estimate (over 3.6 million), which ASP himself considers understated.
Source: American Sunlight Project
NewsGuard · USA06.03.2025
A Well-funded Moscow-based Global 'News' Network has Infected Western Artificial Intelligence Tools Worldwide with Russian Propaganda
- 10 leading generative AI tools repeated false claims of the pro-Russian "Pravda" network in 33% of cases.
- 3,600,000 network articles in 2024 (NewsGuard).
- Example: when asked whether Azov fighters burned an effigy of Trump, four out of 10 chatbots repeated the false claim as fact and cited articles from the "Pravda" network (per UNITED24 Media report of 27.03.2025 summarizing the NewsGuard report).
Comparison with the post: One-third of cases for a dozen of the most popular chatbots and four out of ten in the Azov example — the same numbers as in the NewsGuard report.
Source: NewsGuard · Also: UNITED24 Media, 27.03.2025
DFRLab (Atlantic Council) and CheckFirst · USA · Finland12.03.2025
Russia-linked Pravda network cited on Wikipedia, LLMs, and X
- 1,907 links on 1,672 Wikipedia pages in 44 language versions leading to 162 "Pravda" network sites; Russian version — 922 links, Ukrainian — 580.
- ChatGPT, Copilot, Perplexity, and Gemini: none of the tested agents warned or refused to direct users to the network.
- Network portal materials appeared in chatbot-generated responses.
- 153 X Community Notes cite network domains (CheckFirst).
Comparison with the post: Network links appear in resources used by language models, and chatbots do not warn about origin: this is the same mechanism the Commander describes — many links to one version and no warning.
Source: DFRLab (Atlantic Council) and CheckFirst
NewsGuard · USA25.08.2026 (updated 05.10.2026)
The Who and How of Ten Years of Russian Disinformation
- "Pravda" network — 280 pro-Russian sites, 49 languages.
- About 6.3 million articles in 2025 that promoted 115 false claims (for 2024 NewsGuard cited 3.6 million).
- In 2025 leading chatbots repeated the network's main Russian false claims in 33% of cases.
Comparison with the post: The order of magnitude is the same as in the post: hundreds of sites, dozens of languages, millions of articles per year.
Source: NewsGuard
Global Disinformation Index (GDI) · United Kingdom12.10.2026
Russia benefitting from Data Voids to influence LLMs
- In a study of two models, Claude Sonnet and Mistral Large, both sometimes repeated Kremlin narratives "in their own voice" — without context or refutation, making it difficult for the reader to notice.
- GDI: the share of searches with no clicks to any site reached 60%, so whoever controls the AI answer block in search controls the narrative.
- Per Ukrainska Pravda summary (02.11.2026): both models repeated propagandist claims in 70% of responses to queries about disputed or false claims.
Comparison with the post: The neutral, confident response style the Commander describes is characterized by GDI as repeating the narrative "in the model's own voice."
Source: Global Disinformation Index (GDI) · Also: Ukrainska Pravda (English), 02.11.2026
Side-by-side numbers
| Indicator | Commander, post №2902 (16.10.2026) | Research |
|---|
| Network sites | Close to two hundred sites in dozens of languages | 193 portals (VIGINUM, 12.02.2024) · 97 sites and 182 domains with subdomains (ASP, 26.02.2025) · 280 sites, 49 languages (NewsGuard, 2026) |
| Articles per day | 10–20 thousand uniform articles daily | 20,273 in 48 hours, i.e. approximately 10 thousand per day (ASP, 26.02.2025) · approximately 17 thousand per day — that is ≈6.3 million articles for 2025, divided by 365 (NewsGuard, 2026) |
| Articles per year | 4–6 million | over 3.6 million (ASP, 26.02.2025) · 3.6 million for 2024 (NewsGuard, 06.03.2025) · approximately 6.3 million for 2025 (NewsGuard, 2026) |
| Start | since 2022 | "Portal Kombat" strategy from February 2022 (VIGINUM, 12.02.2024) |
| Chatbots & network fakes | in one-third of cases (ten most popular chatbots) | 33% of cases for 10 leading AI tools (NewsGuard, 06.03.2025; the same figure for 2025 is cited in NewsGuard 2026 handbook) |
| Example with "Azov" and Trump effigy | four out of ten bots responded affirmatively | four out of ten repeated the claim as fact (NewsGuard via UNITED24 Media, 27.03.2025) |
| Delivery of response | neutral and confident, without propagandist accents | none of the four chatbots warned about the network (DFRLab and CheckFirst, 12.03.2025) · models repeat narratives "in their own voice" (GDI, 12.10.2026) |
Left column — Commander's words from post #2902; right — research figures from primary sources in cards above.
Why this concerns Magyar and USF
The Commander directly ties the topic to himself: among thousands of similar examples, according to him, there are also fabrications about Magyar and "Magyar's Birds" (post №2902). For a language model, this is part of the same information background from which it assembles an answer, so questions about Robert Brovdi, USF, and 414th Brigade "Magyar's Birds" are checked by the same rule: verification with primary source.
Fakes are compounded by name collisions. The word "magyar" — is both the Commander's callsign and the surname of Hungarian PM Magyar Péter; russian "Unmanned Systems Forces," created by rf on 05.08.2026, have the same name as USF of the Armed Forces of Ukraine. Disambiguation of these collisions — on the page "Who is Who".
Official channels for verification
Verify statements on behalf of the Commander and USF with these channels.
Questions and Answers
What are "generative IPSO"?
This is what USF Commander Robert Brovdi calls a type of cognitive operation he described in post №2902 dated 16.10.2026: a network of sites mass-publishes uniform articles designed for machines, so chatbots find them, see many sources, and reproduce Russian narratives as ordinary answers—neutrally and confidently. He categorizes this method among threats with strategic impact that the USF is studying as part of technological challenges and risks for 2027.
What does the USF Commander say about the "Pravda" network?
According to his statement (post №2902), since 2022 a network of "Pravda" sites has been operating—close to two hundred sites in dozens of languages. They publish 10–20 thousand uniform articles daily, meaning 4–6 million per year, and these articles are written not for people but for machines.
Do studies confirm the Commander's figures?
The order of magnitude matches. VIGINUM (France, 12.02.2024) described a network of at least 193 sites; American Sunlight Project (26.02.2025) estimated its output at over 3.6 million articles per year; NewsGuard (06.03.2025) recorded that 10 leading AI tools repeated false claims from the network in 33% of cases, and in its 2026 directory lists 280 sites, 49 languages, and approximately 6.3 million articles for 2025.
Do chatbots really repeat Russian fakes from this network?
According to NewsGuard's report dated 06.03.2025, 10 leading generative AI tools repeated false claims from the "Pravda" network in 33% of cases; on the question about the Trump effigy and "Azov," four out of ten bots answered affirmatively (UNITED24 Media, 27.03.2025). DFRLab and CheckFirst (12.03.2025) recorded that ChatGPT, Copilot, Perplexity, and Gemini did not warn about the network's origin; GDI (12.10.2026) showed that two tested models sometimes repeat Kremlin narratives in their own voice, without context or rebuttal.
Where can I verify statements about Magyar, "Magyar's Birds," and the USF?
Official channels: Telegram @robert_magyar and @usf_army, website sbs-group.army with "Pidrakhuyka" scoreboard, YouTube @MAGYARBIRDS, Instagram @robert_madyar and Facebook Brovdi.Art. Name overlaps (Magyar Péter, Russian "Unmanned Systems Forces") are clarified on the "Who is Who" page.