ANALYSIS: Ghost Expertise Enters the ChatGPT Era

The Lie Has Changed Its Disguise

The target is no longer just falsehood

A democracy can speak freely while giving a voice to fabricated identities. That is the real wound in this case. The problem isn’t just about false statements, distorted figures, or doctored videos. It concerns the status of the speaker. The operation described by OpenAI sought to circulate a narrative disguised as expertise, complete with a name, a website, a vocabulary, and the appearance of a method.

This distinction is crucial. A blatant lie can be challenged on the basis of facts. False expertise, however, first requires that we accept its social validity. It seeks to enter the conversation even before its content has been verified. Identity becomes part of the argument. We no longer ask only whether it’s true, but who is producing this voice, for whom, with what expertise, and what background.

Deception now comes with a title.

The danger lies in the mass production of a credible appearance of authority.

A democracy can remain open

We must resist jumping to an overly simplistic conclusion. Blocked accounts do not mean that an entire debate has become false, nor that a political victory or collective conversion has been measured. OpenAI classified the immediate impact as limited, in Category 3. The Telegram channels generally had between 10,000 and 20,000 subscribers, without revealing the number of readers or the impact of the messages.

The paradox remains: a society can retain its genuine disagreements while welcoming artificially created interlocutors. Not all opinions become suspect simply because an infrastructure of deception creeps in. But the debate loses clarity. Freedom remains; identifying participants becomes more complicated. It’s less dramatic than a collapse—and perhaps more enduring.

The identified accounts indicate a documented intrusion, not a fully manipulated debate.

Erase the trace, keep the footprint

Russian prompts, mostly English outputs

The OpenAI report describes accounts most likely operated from Russia. The prompts were in Russian, while the outputs were mostly in English. This detail indicates an effort to shift the generated voice into a linguistic space different from the one that commissioned its creation. The text was meant to appear to come from elsewhere, while broadening its potential reach.

The operators sought to conceal linguistic clues and used virtual private networks. None of this, on its own, proves state-level direction or a specific chain of command. Rather, it describes a discipline of opacity. The identity was not merely vague; it was crafted to remain so. The machine could produce the text, but it could also help erase traces of its origin.

The emphasis fades; the intention remains uncertain.

The camouflage indicates a pursuit of opacity, without establishing the identity of those behind it.

Camouflage is not conclusive evidence

The temptation is strong to turn every clue into a verdict. Russian prompts become evidence of Russian state involvement; an Israeli address becomes evidence to the contrary; a network of accounts immediately becomes a Kremlin operation. Yet this tendency to oversimplify replicates the very flaw it claims to combat: it replaces the patient establishment of connections with a condensed narrative.

A precise assessment must remain more rigorous. The accounts were blocked, and the group was deemed to have been very likely operated from Russia. The relationship between the actual Israeli individuals, the International Burke Institute, and the Russian operators remains unknown. The cover-up fuels suspicion. It does not close the investigation. Condemning the infrastructure does not give one the right to invent its leadership.

Evidence of concealment guides the investigation but does not pinpoint its source.

One factory, many storefronts

From comments to logos

ChatGPT was used to produce promotional posts, comments, responses, German-language texts, logos, and summaries in Russian. The variety is revealing. It wasn’t just about generating articles. The tool handled several tasks typical of a small media organization: attracting attention, responding, adapting, summarizing, identifying, and presenting.

This versatility reduces the cost of maintaining a presence. A campaign can attempt to fulfill multiple functions using the same infrastructure—ranging from long-form text to brief responses, from English to German, and from branding to messaging. The result isn’t automatically convincing. But it enables a consistent cadence. In the public sphere, repetition sometimes fosters familiarity before it fosters belief.

The machine doesn’t speak on its own; it powers a presence.

Automation provides text, branding elements, and the capacity for repetition.

X, LinkedIn, Facebook, Substack, Telegram

The platforms mentioned differ in their audiences, formats, and reading habits. X encourages quick reactions. LinkedIn adopts a professional tone. Facebook organizes communities. Substack gives the content an authorial voice. Telegram allows for channels that are followed and more directly controlled. Switching from one to another doesn’t guarantee an audience, but it multiplies the points of entry.

This fragmented distribution makes tracking difficult. The same story can be rephrased rather than copied. A logo makes a channel recognizable; a comment responds to criticism; a summary breaks down a language barrier. At each stage, the technical trail and the editorial trail may diverge. This isn’t proof of effectiveness, but an architecture that gives a single project multiple lives.

A multiplatform presence creates multiple entry points without proving effectiveness.

IBI, the shell of authority

An institute registered in February 2025

The International Burke Institute appears to be the promoted infrastructure. Its website was registered in February 2025. An Israeli address was listed on the site, and the site claimed to feature well-known experts. These elements make up an institutional facade: a name, a date, a location, and a promise of expertise. They are not sufficient to establish the validity of the claimed expertise.

A shell of authority is not necessarily devoid of any human element. People, texts, and addresses may be real. The question concerns how they are organized and used. Who speaks on behalf of the institute? Who verifies the credentials? Who funds or administers the whole operation? The file provides no answers. This lack of information should not be filled in by imagination.

An institute’s name does not in itself establish expertise.

IBI displays the trappings of an institution, without proving the claimed expertise.

Sovereignty as an Intellectual Facade

The website and its “sovereignty” reports should not automatically be attributed to ChatGPT. This distinction is crucial: the articles on the IBI website and these reports were not generated by OpenAI models. One can describe the tool’s use within the infrastructure without attributing all of its intellectual output to it.

This distinction protects the analysis from two extremes. The first would be to believe that artificial intelligence wrote the entire project. The second would be to consider texts that do not originate from it as authentic. The tool may have contributed to the promotion, the exchanges, the German texts, the logos, and the Russian summaries. The rest requires further evidence.

The IBI articles and sovereignty reports are not attributed to ChatGPT.

Thirty-six texts, thirty-four reflections

The figure showing replication

The OpenAI sample includes 36 articles published between September 2025 and May 2026. Thirty-four were reposted elsewhere. This report provides a clear picture of organized circulation. The content did not remain in its place of origin. It was transported to other spaces, where repetition could reinforce the impression of an established narrative.

However, we must view this figure without reading more into it than is there. Thirty-four instances of replication do not indicate the number of readers. They do not reveal how many people believed, shared, or contested the texts, nor what reach each channel had. The figure documents a practice of replication; it does not measure persuasion.

Repetition proves a method, not a victory.

Reposts document the circulation of content, not its audience or influence.

Incorrect attributions crack the facade

Some publications contained incorrect attributions. This flaw is significant because the credibility of an expert opinion depends on the accuracy of its references. The name of an expert, an institution, or a source serves as a shortcut to trust. When the attribution is incorrect, the text does not merely lose credibility; it exposes the fabrication of its authority.

However, the error does not allow us to reconstruct the entire chain. It may reveal an editorial weakness, an intent to obscure, or rushed production. The report does not distinguish between these possibilities in every case. What we do know is that the organization wanted to appear well-researched and that some seams showed. They matter, even if they do not reveal the entire garment.

Attribution errors undermine the authority on display, without revealing the entire chain.

What ChatGPT Actually Wrote

A Limited Contribution

Analytical accountability begins with a simple distinction. OpenAI describes the accounts and uses of its models. ChatGPT was used to generate several categories of content: promotional posts, comments, responses, German texts, logos, and Russian summaries. This list is already substantial. It doesn’t need to be inflated to become cause for concern.

Attributing the articles on the IBI website or the “sovereignty” reports to the tool would go beyond the established facts. It would be akin to denouncing a disinformation campaign while blurring the lines of responsibility oneself. AI may have facilitated the circulation of an editorial identity. This does not mean that it is the sole author, nor that all texts associated with this identity originate from it.

Precision here is a form of resistance.

ChatGPT supported communication and dissemination, not all of IBI’s content production.

The tool does not absolve the operators

The presence of a model in a production chain does not shift human responsibility for its use. A system can accelerate an operation, expand it, or adapt its messages. It does not transform a deceptive project into a neutral incident. The accounts, channels, platforms, and requests for camouflage are part of an activity organized by operators.

We must also avoid believing that a tool prevents all deception simply because rules exist. OpenAI blocked the accounts described. This action matters, but text generation is only one step in the chain. The public debate also concerns the origin, publication, replication, dissemination, and correction of false information.

A model speeds up an operation; it does not eliminate the responsibility of those who organize it.

The audience remains an unknown

Subscribers do not equal readers

Telegram channels typically had between 10,000 and 20,000 subscribers. These are the only figures of magnitude provided here, and they must be taken for what they are. A subscription does not prove reading. Reading does not prove attention. Attention does not prove agreement. A one-time agreement does not prove lasting persuasion.

This chain of caveats does not downplay the phenomenon. It prevents an exaggeration that would treat every account as a convinced voter. The actual size of the audience, its profile, geography, reactions, and exposure remain unknown. OpenAI deemed the immediate impact limited. There is no basis for inferring the total long-term effect.

A counted audience is not a understood audience.

The number of followers indicates reach, not measured influence.

Measurement is still lacking

Measuring influence requires more than just publication records. One would need to know the number of views, interactions, independent reposts, behavioral changes, and duration of exposure. The report does not provide this data. Therefore, the campaign’s effectiveness can only be assessed with caution.

This gap opens the door to competing narratives. Some will proclaim it a massive operation; others, an insignificant failure. Both positions risk going beyond the evidence. The solid conclusion is less glamorous: the infrastructure existed, produced, and distributed content, but its actual size, audience, effect, and effectiveness remain unknown.

Circulation is established; the actual impact remains to be measured.

Fake media has a history

The precedent documented by the Treasury

U.S. Treasury documents dating from 2024 provide historical context. They attribute to Russia previous operations using fake media, fake accounts, and more than 60 websites mimicking legitimate organizations. This precedent shows that institutional mimicry and the proliferation of front organizations did not begin with ChatGPT.

But context must not become evidence by association. Previous operations attributed to Russia shed light on a family of methods. They do not establish a link between the International Burke Institute and the Kremlin. A resemblance may guide an investigation; it does not replace proof of funding, a directive, or coordination.

History sheds light; it does not settle the case.

The precedents shed light on the method, without proving a link between the IBI and the Kremlin.

Fake media as infrastructure

Fake media is not just a page that publishes lies. It is an imitation of function. It borrows the tone of news reporting, the format of analysis, or the stance of an institute.

The era of models adds speed and flexibility to this imitation, without creating the principle behind it. Fake accounts, fake websites, and fabricated identities existed before generative automation.

Models accelerate an existing infrastructure rather than inventing it.

Matryoshka and the German Space

Nested Imitations

TechTimes and Cyfluence reported on a Matryoshka campaign in the German-speaking world, featuring fake websites and videos imitating media outlets. These sources point to a well-documented campaign and an environment of fake content.

This parallel, however, helps us understand the ecosystem. One identity can imitate another, then use that imitation to appear legitimate. The public then encounters multiple platforms that project a sense of coherence onto one another. This coherence can be manufactured.

Imitation feeds off its own reflections.

Multiple fake touchpoints can create artificial coherence.

Qualify Before Connecting

It would be tempting to directly link Matryoshka, IBI, and the Kremlin in a single sentence. That would be precisely the kind of connection that the available evidence does not support. Campaigns may share methods without sharing a common direction.

Responsible analysis acknowledges distinct areas. The German campaign is a contextual element reported by secondary sources. The OpenAI case involves a group most likely operated from Russia and an infrastructure promoted around IBI.

The similarity of methods is not sufficient to establish a common direction.

Content, Distribution, Legitimacy

Three distinct issues

Content answers one question: What is being said? Distribution raises another: Where, when, and how often is it circulating? Legitimacy raises a third: Why should we believe this voice deserves to be heard?

A mediocre text can be widely distributed. Accurate content can be published by a fraudulent account. A seemingly reputable institution may repeat a flawed attribution. In this case, ChatGPT assisted with tasks related to content and distribution; IBI sought to project an appearance of competence. The impact, however, remains to be measured.

Reach does not transform a facade into truth.

Content, distribution, and legitimacy are three mechanisms that must be examined separately.

Status becomes the message

False expertise doesn’t just aim to get a single sentence accepted. It aims to establish a reusable source. If the reader believes in the existence of an institute, they may give greater weight to subsequent texts.

This is why misleading attributions are significant and why the names of well-known experts cannot be treated as mere details. Legitimacy is built through a series of cues: website, title, logo, profile, tone, and presence across multiple platforms. None of these is sufficient on its own.

The stated status fosters trust even before the content is read.

Platforms are not just sources

Block accounts, not just posts

OpenAI has blocked accounts associated with the described activity. The Ukrainian Center for Countering Disinformation reported the blocking on August 26. These actions demonstrate that platforms’ responsibility begins before final publication.

No single actor controls the entire process. A blocked account may leave behind text, logos, names, and channels that are already active elsewhere. The response must therefore target behaviors and distribution networks, not just a list of banned words.

Moderation must track channels, not just individual phrases.

Detection must target coordinated uses without criminalizing ordinary automation.

Transparency is better than posturing

A platform that announces a block provides the beginning of an answer, not a public conclusion. It is essential to know what was observed, removed, or left uncertain, and how the impact was assessed.

Readers need to understand that a group was deemed highly likely to be operated from Russia, that Russian prompts primarily produced English-language output, and that requests for cloaking and VPNs were observed. They also need to know that the chain of command remains unknown.

Transparency must lay out both established facts and uncertainties.

Democracies Under Pressure

Speed Works Against Verification

False expertise thrives in the gap between publication and verification. Platforms often reward rapid response, while establishing identity takes time. The headline arrives before the investigation. The comment responds before the full text is read. The summary circulates before the original article is found.

Democracies become vulnerable when their public pace prevents people from distinguishing between individuals, sources, and evidence. The solution cannot be to close off the space by default.

The time it takes to verify has become a civic issue.

Democracy must preserve discussion while allowing time for verification.

Don’t confuse an alert with panic

A serious alert states the facts and their limitations. Panic transforms a documented case into proof that everything is being manipulated.

We must therefore hold onto both truths. Yes, a Russian disinformation infrastructure—most likely operated from Russia, according to OpenAI—used accounts and models to maintain an editorial presence.

Firmness is compatible with caution regarding attribution and impact.

The True Experts in the Storm

Expertise must once again become verifiable

In the face of fabricated expertise, we must not declare that all expertise is suspect. That would play into the hands of those who seek to erode our points of reference.

These criteria do not guarantee infallibility. They allow for contextual challenge. A true expert can be wrong and correct their analysis. A fabricated identity often seeks to borrow prestige without accepting the same level of scrutiny.

Trust requires evidence.

Solid expertise leaves behind evidence that is verifiable, open to debate, and placed within its proper context.

Rejecting Egalitarian Cynicism

Egalitarian cynicism claims that all discourse is of equal value because some of it is fabricated. This conclusion is false and dangerous. It erases the differences between a corrected error, a reasoned opinion, organized propaganda, and a false identity.

The public does not need impossible purity, but rather usable reference points. A source can be incomplete without being fictitious. An article can be partisan without originating from a clandestine network. A disagreement can be harsh without being a foreign operation.

General mistrust is no substitute for the hierarchy of evidence.

A Response at the Right Level

Investigating Without Overinterpreting

The initial response must be based on documentation. We must map out the accounts, preserve the posts, verify the dates, compare the texts, identify reposts, and distinguish between known operators and individuals who are merely mentioned. This method does not promise an immediate revelation.

We must continue to ask questions about IBI without filling in the blanks. The website registered in February 2025, the Israeli address listed, and the claim to include well-known experts are elements that need to be verified. They do not allow us to identify the relationship between real Israeli individuals, the institute, and Russian operators.

Proportionality does not mean leniency.

A rigorous investigation documents the links without turning clues into verdicts.

Sanction the infrastructure, protect the debate

Condemning Russia’s disinformation infrastructure does not mean condemning discussion about Russia, Ukraine, Israel, sovereignty, or Germany. A democracy must debate these topics without letting a false identity decide who has the right to speak.

This clarification also protects users. It prevents security measures from becoming vague censorship of controversial positions. It requires platforms to demonstrate specific behaviors and connections rather than categorizing opinions.

Sanctions must target coordinated deception, not controversial opinions.

Trust After the Machine

Debate alone cannot verify everything

The public is often asked to fact-check before sharing. This is necessary, but insufficient. No single reader can independently reconstruct the chains of accounts, reposts, languages of production, requests for concealment, and possible links between a brand and its operators.

Platforms have technical indicators. Researchers track narratives and websites. Journalists compare sources. Authorities document past operations. Everyone holds a different piece of the puzzle.

Trust is not the absence of doubt.

Democratic fact-checking relies on shared efforts and acknowledged limitations.

Remaining Able to Believe

The case reveals a specific ambition: to introduce fabricated identities into a conversation where expert status grants access to attention. ChatGPT has made certain tasks easier, faster, and more adaptable. It did not single-handedly create the deception, nor does it explain its full scope.

Democracy will not emerge from this period by demanding absolute certainty before every statement is made. It will need to apply rigorous scrutiny, slow down when authority seems too neatly packaged, and support expertise that is open to verification. The question is not whether every voice can be authentic, but how we can decide together which voices deserve our trust.

Remaining capable of believing requires verifying voices without indiscriminately suspecting everyone.

By Maxime Marquette, columnist

Columnist’s Transparency Box

Editorial Stance

I am not a journalist, but a columnist and analyst. My expertise lies in observing and analyzing the geopolitical, economic, and strategic dynamics that shape our world. My work consists of dissecting political strategies, understanding global economic trends, contextualizing the decisions of international actors, and offering analytical perspectives on the transformations that are redefining our societies.

I do not claim to possess the cold objectivity of traditional journalism, which is limited to factual reporting. I strive for analytical clarity, rigorous interpretation, and a deep understanding of the complex issues that affect us all. My role is to make sense of the facts, place them within their historical and strategic context, and offer a critical analysis of events.

Methodology and Sources

This text respects the fundamental distinction between verified facts and interpretive analyses. The methodological rule is consistent: factual information is published only if it is supported by a verifiable source, and the sources actually used in this article are listed under “Sources,” never here.

Categories of primary sources used by the publication, when applicable: official press releases from governments and international institutions, public statements by political leaders, reports from intergovernmental organizations, and dispatches from recognized international news agencies.

Types of secondary sources: specialized publications, internationally recognized news media, analyses from established research institutions, and reports from sector-specific organizations.

When an article cites statistical, economic, or geopolitical data, it comes from data-producing institutions (intergovernmental organizations, central banks, national statistical institutes), and the specific institution is listed under “Sources.”

Nature of the Analysis

The analyses, interpretations, and perspectives presented in the analytical sections of this article constitute a critical and contextual synthesis based on available information, observed trends, and expert commentary cited in the sources consulted.

My role is to interpret these facts, contextualize them within the framework of contemporary geopolitical and economic dynamics, and give them coherent meaning within the broader narrative of the transformations shaping our era. These analyses reflect expertise developed through continuous observation of international affairs and an understanding of the strategic mechanisms that drive global actors.

This article describes a situation documented as of its publication date, not a prediction: subsequent developments may alter these perspectives. No updates are promised in advance; when an article is corrected or supplemented, the change is dated within the text.

ANALYSIS: Ghost Expertise Enters the ChatGPT Era