“A hacker group called Black Mirror has published screenshots of an analytical note from Sberbank that lists eight scenarios for the end of the large war between Russia and Ukraine.”
A wonderful story, and it does illustrate our reality. What it illustrates belongs to the intellectual field. It has little to do with the war or with politics.
The “analytical note” from Sberbank, supposedly obtained by some hackers, turns out on closer reading to be an AI-generated text, with errors and hallucinations.
I have heard that someone rated the AI generation at 70 percent. My own run gave at least 90 percent.
So if we assume that the texts really do come from Sberbank, the conclusions are sad. Behind the text with its eight scenarios for the end of the war stands a poorly prepared person, or a group of people, who wrote a worthless prompt and whose output is presented as the boiling analytical thought of some unknown minds on Gref’s team.
Since I produce analytical AI output myself every day, I can “state with full responsibility” that people who draw a salary at Sber for a result like this should be thrown out.
American LLMs (I have not tried the Chinese ones, but the situation there is probably similar) perceive the Russian state and Russian society through their own assumptions about a normal Weberian state, or at best as an authoritarian regime with a classic set of characteristics.
The Russian Federation as a state formation is such a puzzling phenomenon that there is still no intellectual language for describing it. There is no scientific and terminological apparatus that would allow an adequate description of the mechanisms by which political power is exercised, or of so-called “Russian society”.
Every political thought spoken about Russia is a lie, for the reasons given above. Any generalization about Russia and about “Putin” (that is, about political power in the Russian Federation) is a lie, if we approach the question with epistemic honesty. This includes, of course, the author of these lines.
As for LLMs, their narrative is a lie twice over, or more precisely a multiplication of lies, since their analysis rests on a countless number of clichés and prejudices about the present-day Russian Federation and its system of political power.
Moreover, the efforts of the top group of Putin’s managerial class are aimed at actively hiding any real mechanisms of rule and coordination, and at hiding any real contacts and communication with the outside world, the kind that is not staged.
There is a very harsh sanitary regime, a regime of elimination of any potential source of information about the real mechanisms and the real political actors of power in the Russian Federation.
For AI models to take this insanely complex and unclarified context into account, they need long training, with no guarantee that it will actually succeed.
So the situation in the field of political analysis of the present-day Russian Federation, and of political power in it, requires two things. First, cutting off the prejudices and pseudo-knowledge about Russia that burden almost all Euro-Atlantic analytical thought without exception. Second, laborious and long refinement of AI tools on this completely non-standard object.
In this light the experience of Russian managers using “AI analytics” themselves is interesting, in this case no-names from Sberbank (Gref, nominally).
Here the use of AI leads them into a trap of pseudo-analytics from untrained, or badly or wrongly trained, LLMs, which carries all the templates and stereotypes about the structure of power and society in Russia.
Addendum 29.08.2026
A few days after the publication of the so-called Sberbank “analytical note”, the same channel, Black Mirror, clarified the details. This is not a corporate product. It is the work of German Gref himself, or more precisely of the AI model for which he or someone else wrote the prompt. The Sberbank pseudo-analytics turned out to be part of the personal correspondence between Gref and Metropolitan Tikhon (Shevkunov), whom analytical discourse calls Putin’s confessor.
.
Leave a Reply