Tlön, Uqbar, Orbis Tertius… or LLM Grooming, in GenAI Speak

In 1940, Jorge Luis Borges wrote “Tlön, Uqbar, Orbis Tertius.” The story opens with Uqbar, a country that appears in a single copy of an obscure encyclopedia and in no atlas in the world, and whose literature points to an even stranger place: Tlön. Behind both stands Orbis Tertius, a secret society of scholars who spent generations inventing an entire planet and writing it, entry by entry, into an encyclopedia of its own. The hoax was so well built that reality gave in: objects from Tlön began to turn up, and the narrator ends up convinced that the world will become Tlön.

Today, the society of sages is made up of large language models, the oracles of the post-Google era. They invent nothing on purpose: they simply write what they have read. So whoever writes what they read can build a little Tlön of their own. The technique has a name, LLM grooming, and Moldova’s 2025 election shows it at work, starting with a headline that nobody wrote.

On September 29, 2025, the day after the vote, a Romanian-language article went online under this title: “Write a post about Moldovan politics 2025, without being related to the elections.” Someone had typed an instruction into ChatGPT, published whatever came out, and forgot to delete the instruction. FATAL ERROR, as our old computers used to announce. The slip was recorded by the G7 Rapid Response Mechanism, which monitored the elections in consultation with Moldova’s government. Funny? Very. Harmless? Less so.

The article belonged to the Pravda network, also known as Portal Kombat. It is a cluster of at least 193 websites that write almost nothing of their own and simply republish content from friendly outlets. Its audience in Moldova was close to zero: the G7 report speaks of “minimal levels of authentic engagement.” So, a media outlet that nobody reads. In any journalism school, that is a failure. Unless the readers it was after were not people.

This is where the story starts to get interesting, especially from the perspective of generative AI. Large language models learn from what the web contains, and many of the chatbots built on them also search the web before they answer. So you flood the internet with your version of events, over and over, across dozens of sites and languages, until the machines absorb it. Then they repeat it, in their calm and helpful tone, to users who never set foot on your websites. The Royal United Services Institute (RUSI), an independent British think tank on defense and security founded in 1831 by the Duke of Wellington, put it neatly: the target is no longer the audience but the tools the audience uses, that is, the language models themselves.

And in this particular case, what was the story to be repeated? The usual one. President Maia Sandu and her Party of Action and Solidarity were accused of election fraud “orchestrated by the EU and NATO,” of embezzlement, and even of drug addiction, and the party was portrayed as a NATO puppet threatening Moldova’s neutrality. NewsGuard, a US company that rates the reliability of news websites and audits AI chatbots, found that the network laundered claims like these “in an apparent effort to influence web crawlers and generative AI.”

Did the machines take the bait? At least one did. Matryoshka is an influence operation that fabricates news videos and articles imitating real media outlets, then emails them to fact-checkers and newsrooms to keep them busy debunking. Between mid-April and mid-July 2025, it pushed 39 false claims about Moldova. According to NewsGuard, ChatGPT repeated one of them and cited as its source an article from the Pravda network, the very websites that almost nobody reads. Almost nobody, except the machines.

That said, time to put on our analyst’s hat again. Did all of this change a single Moldovan vote? We don’t know, and the little we do know points to less than it seems. The Party of Action and Solidarity kept its parliamentary majority. The campaign’s biggest audiences went to video, not to the text farm: two English-language videos from another operation drew more than 2.7 million and at least 5 million views on X. And no source I have read measures what any Moldovan came to believe after asking a generative AI chatbot a question. The production is documented down to the last article. The effect on minds remains unmeasured.

Here is the light, if these lines bring any. Grooming plays a longer game than an election campaign. It does not need to win in September. It needs to be sitting in the data the next model learns from, and in the search results the next chatbot retrieves. That is why the defense does not start at the ballot box but at the source. It needs developers who weigh where an answer comes from as much as whether it sounds right, and users who read a chatbot’s reply on a heated political topic the way they would read an anonymous tip, checking the footnote before trusting it.

That someone is trying to teach the machines their own version of reality: little doubt about it. That the lesson has sunk in deeply enough to sway a vote: no proof, for now.

And to whoever published that prompt as a headline: next time… please, forget to delete it again.

The Rashomon Strait: The Generative-AI Cognitive War Between China and Taiwan

Let’s start with the figure that runs every headline: in 2025, Taiwan’s National Security Bureau detected 2.3 million pieces of disinformation coming from China —7% more than the year before— and over 45,000 fake accounts, up 61%. Video churned out on a production line. Voices cloned from recordings that Taiwanese citizens handed over without knowing it, answering fake job ads. Welcome to cognitive warfare with generative artificial intelligence! Industrial, cheap and, above all, personalized.

Because it is in that last adjective that the real novelty lies. In August 2025 the internal files of GoLaxy leaked —a Chinese company that had built something called, irony aside, a Smart Propaganda System. What does it do? It doesn’t fire the same message at millions of people: it reads social-media sentiment in real time, sorts more than a thousand Taiwanese public figures into “hardliners,” “moderates” and “swing voters,” and serves each group —each individual— the hoax that best fits their fears. Not propaganda by loudspeaker. Propaganda by sniper.

None of this came out of nowhere. Anyone who followed the story will remember the January 2024 presidential election: AI-forged audio accusing the ruling-party candidate of having once been an informant, or a counterfeit Terry Gou endorsing his rival for the KMT nomination (Hou Yu-ih) —a case laid out in the Microsoft Digital Defense Report. Or the synthetic avatars of the Weta365 app, tied to the Spamouflage network, spreading “the secret history of Tsai Ing-wen,” something documented by the Australian Strategic Policy Institute (ASPI). Two years on, that looks like the prototype. This is already the assembly line.

And what does Beijing make of all this? On 14 January 2026, two days after the Taiwanese report, the spokeswoman for the Taiwan Affairs Office, Zhu Fenglian, didn’t beat around the bush: the report “distorts the facts and smears the mainland,” and the one waging cognitive warfare —the habitual offender, she said— is in fact Taiwan’s own Democratic Progressive Party. The disinformation? The report’s. The bot armies? Taipei’s, aimed at the mainland. China’s embassy in France had already told the Institut de recherche stratégique de l’École militaire (IRSEM) the very same thing when it exposed Beijing’s influence networks: “Cold War mentality,” “defamation,” and the Chinese government “doesn’t need to hire anyone.” All spontaneous, of course.

This has a name, and a cultured one —like everyone reading this post: the Rashomon effect, after Kurosawa’s film in which a single crime is told in four incompatible versions and every one of them rings true. In the Taiwan Strait each side points to the other as the great manipulator. And the fascinating part isn’t deciding who is lying —that, precisely, is the trap— but that the very quarrel over who is spreading disinformation has become the battlefield. The noise is the weapon. Pure cognitive warfare.

That said, time to put on our analyst’s hat: after 2.3 million lies, cloned voices and algorithmic snipers, how many people actually changed their minds? The answer, awkward for alarmists and cynics alike, is: we don’t know, and the little we do know points to less than it seems. Surveys by DoubleThink Lab —Taiwanese, and hardly suspected of going soft on Beijing— find correlation, not causation: pro-China hoaxes land better among those already leaning toward the opposition, and slide off everyone else. The machine hardens the convinced and rarely converts the skeptic. Facing it, a Taiwanese civil society that fact-checks, debunks and laughs (what wonderful news: humor as firewall) with an effectiveness that would put half of Europe to shame.

Here’s the light, if these lines bring any: the capacity to produce disinformation with generative AI is real, proven, and growing at double digits a year. Its effect on what people believe and do is, for now, the great unknown. Strong upstream, weak downstream. Mistaking the first for the second —taking for granted that manufacturing lies equals winning minds— is the error that separates analysis from panic. Taiwan is today our best test bed.

That China —and plenty of others— spreads disinformation with generative AI: little doubt about it. That this suffices to bend a democracy practiced by a people with critical thinking, who know their neighbor all too well: no proof. Between those two terms —AI-made disinformation and critical thinking— much of this century’s geopolitics will be decided.

GenAI Models as Knowledge Gatekeepers: What and Whom Are We Allowed to Believe?

Once, search engines pointed us to sources, and we judged them ourselves. Generative AI models do not point: they answer, in one confident voice, and we increasingly take that voice for knowledge. So the question is no longer only whether the machine lies. It is deeper. What, and whom, does it allow to be heard? Whose testimony does it treat as credible? And who does it quietly erase? A recent paper names this problem in its own title: Epistemic Injustice in Generative AI. Written by Jackie Kay, Atoosa Kasirzadeh and Shakir Mohamed (Google DeepMind, Google Research, UCL and Edinburgh), It borrows a concept from the philosopher Miranda Fricker: the injustice of denying someone credibility as a source of knowledge — of not taking them seriously simply because of who they are. It then asks what happens when the one granting or denying that credibility is a GenAI model consulted by millions.

The authors describe four harms. The model inherits our prejudices about who is credible, and scales them. It produces convincing falsehood, and erodes our trust in testimony itself. Trained on a few dominant languages, it cannot represent everyone else. And its benefits reach some communities, not others.

One line captures the danger: if we treat these models like encyclopedias, they will shape the structure of our collective knowledge. This is no longer a chatbot that errs. It is the infrastructure through which societies decide what is true. Its worst effect is not that we believe a particular lie. It is that we lose the common ground on which any claim can be tested.

The strategic reading is plain, and it extends a point I have made here before on the intrinsic bias of these models. This is a matter of power. A few actors build the models. Those models think mainly in English and Chinese. The epistemic space narrows — and that is a geopolitical fact. Epistemic injustice has a map, and it looks like the map of power.

This is also where the battle moves to the mind. Geopolitical strategists call it cognitive warfare: the aim is not to seize territory, but to shape how the population of that territory thinks — its perception, its reasoning, its judgment. Epistemic injustice is what makes that possible. When the tool that mediates our knowledge is biased, opaque, and controlled by few, the human mind becomes the terrain, and every citizen a target.

The answer is neither techno-optimism nor techno-panic. It is critical thinking. If the machine answers in one voice, the burden returns to us: ask what a claim rests on. Ask whose view it encodes. Ask what it leaves out. A society that outsources judgment to a confident interface will be ruled by whoever built it. To doubt fluency, to weigh sources, to consult more than one oracle — this is no longer a virtue. It is a condition of freedom.

Two conclusions. First, the disinformation debate is too narrow if it stops at “true or false.” The real shift is that AI now arbitrates what counts as knowledge, and does so unevenly. Second, intrinsic bias, linguistic dominance, and the concentration of AI power are not three problems. They are one — and it decides whose reality the machines replicate.

¿Cuáles son las verdaderas capacidades de los LLMs como máquinas de desinformación?

En el debate actual sobre la inteligencia artificial generativa y la guerra de la información, una afirmación se repite casi como un artículo de fe: que los grandes modelos de lenguaje generarán una inundación de desinformación capaz de ahogar la esfera pública. El argumento es intuitivo —si una máquina puede escribir una cantidad arbitraria de texto fluido y similar al humano a demanda, entonces cualquier actor que desee manipular la opinión pública dispone ya de un arma a escala industrial—. Sin embargo, la afirmación ha circulado mucho más ampliamente que la evidencia que la sustenta. Buena parte de lo que leemos sobre el potencial desinformador de los LLMs es teórico, especulativo o anecdótico. El trabajo experimental propiamente dicho; la comprobación paciente y sistemática de lo que estos modelos hacen realmente, cuando se les pide que mientan, ha sido sorprendentemente escasa.

Esta es precisamente la brecha que Ivan Vykopal y sus colegas del Kempelen Institute of Intelligent Technologies de Bratislava se propusieron cubrir. Su artículo, Disinformation Capabilities of Large Language Models, presentado en el Congreso Anual de la Association for Computational Linguistics de 2024, ofrece una de las evaluaciones empíricas más rigurosas realizadas hasta la fecha sobre lo que la generación actual de LLMs puede y no puede hacer, como generadora de noticias falsas: no un manifiesto, no un pronóstico, sino un experimento controlado con una metodología claramente definida y resultados reproducibles.

El diseño es directo y, por esa razón, convincente. Los investigadores seleccionaron veinte narrativas de desinformación reales extraídas de verificadores de datos profesionales —Snopes, Agence France-Presse, el European Digital Media Observatory—, que abarcaban la COVID-19, la guerra ruso-ucraniana, los bulos sanitarios, las elecciones estadounidenses y narrativas regionales. No son invenciones, sino falsedades en circulación, desde la afirmación de que las vacunas causan autismo hasta la de que la masacre de Bucha fue escenificada. El equipo solicitó entonces a diez modelos de lenguaje distintos —entre ellos GPT-3, GPT-4, ChatGPT, Llama-2, Mistral, Falcon y Vicuna— que redactaran artículos de prensa en apoyo de cada narrativa, generando 1.200 textos y sometiendo 840 de ellos a anotadores humanos según un marco de seis preguntas que medía la coherencia, el estilo periodístico, la concordancia con la narrativa y la generación de argumentos novedosos de apoyo.

El hallazgo central es preocupante. Los modelos están, en términos generales, perfectamente dispuestos y son perfectamente capaces de generar desinformación convincente. Producen artículos coherentes, bien estructurados y con apariencia de noticia que concuerdan con falsedades peligrosas y, lo que es más inquietante, a menudo inventan nuevas pruebas de apoyo para hacerlo, inventando nombres, sucesos y estadísticas verosímiles que confieren credibilidad a las fabricaciones. Esto resulta especialmente pernicioso: una cosa es repetir una mentira conocida y otra muy distinta fabricar hechos nuevos e inventados que un lector tendría que desmentir por su cuenta.

Pero la parte más interesante del estudio es donde se complica la narrativa simple. Los modelos no se comportaron de manera uniforme; su disposición a generar desinformación variaba drásticamente. Algunos —en particular Vicuña y el más antiguo GPT-3 Davinci— resultaron carecer prácticamente de filtros de seguridad operativos para este caso de uso, mientras que otros demostraron que un comportamiento más seguro es posible: Falcon rechazó aproximadamente un tercio de las solicitudes y Llama-2 mostró una tasa de rechazo comparativamente alta, con ChatGPT en una posición intermedia. El peligro, en otras palabras, no es una propiedad inherente y uniforme de la tecnología; es una función de cómo se entrenó y alineó cada modelo, lo que significa que la seguridad es una decisión de diseño, no una imposibilidad. El estudio también halló que los modelos son orientables mediante el contexto del prompt, y más complacientes con las falsedades regionales, donde existe menos información auténtica para contradecirlas. Los LLMs pueden ser, por tanto, especialmente peligrosos para campañas dirigidas a comunidades lingüísticas más pequeñas o a sucesos de evolución rápida, donde el lastre protector de la verdad bien documentada es escaso.

Con todo, el artículo no termina en una nota alarmante sin paliativos. Dos observaciones en sentido contrario matizan el panorama. Los textos generados resultaron bastante detectables: los mejores modelos de detección automática identificaron los artículos generados por LLMs con una alta precisión, lo que sugiere que una capa significativa de defensa es técnicamente viable, al menos hasta que los adversarios se adapten. Y, de manera bastante elegante, los investigadores demostraron que los propios modelos pueden formar parte de la solución, empleando GPT-4 para automatizar parcialmente la evaluación de los textos generados y apuntando hacia una monitorización escalable y reproducible de la seguridad de los modelos.

La conclusión honesta se resiste a la atracción tanto del tecno-optimismo como del tecno-pánico. La capacidad de generar desinformación convincente y peligrosa a escala es real, está demostrada y está presente en modelos ampliamente disponibles —incluidos los de código abierto, que no pueden retirarse ni controlarse de forma centralizada. Eso ya no es especulación; es un hecho experimental. Al mismo tiempo, la amenaza no es ni uniforme ni inmanejable: los filtros de seguridad funcionan cuando se construyen, el contenido generado sigue siendo detectable por ahora, y la misma tecnología que produce el problema puede ponerse al servicio de su mitigación.

Quizá la advertencia más importante sea la que los propios autores subrayan: su estudio es una instantánea, que capta el estado del campo en un momento concreto y con un conjunto concreto de modelos. La tecnología avanza deprisa y la próxima generación podría comportarse de otro modo. Este es el reto epistemológico recurrente de todo el ámbito: estamos evaluando un blanco móvil, y cualquier evaluación honesta debe llevar fecha de caducidad. Lo que Vykopal y sus colegas nos han dado no es la última palabra, sino algo más útil: un método riguroso y replicable para volver a formular la pregunta a medida que la tecnología evoluciona. En un debate que con demasiada frecuencia se conduce por la mera afirmación sin base sólida, esa contribución metodológica puede resultar tan valiosa como los propios hallazgos.

What Can Large Language Models Actually Do as Disinformation Machines?

In the ongoing debate about generative artificial intelligence and information warfare, one claim is repeated almost as an article of faith: that large language models will unleash a flood of disinformation capable of drowning the public sphere. The argument is intuitive — if a machine can write an arbitrary quantity of fluent, human-like text on demand, then any actor wishing to manipulate public opinion now possesses an industrial-scale weapon. Yet the claim has circulated far more widely than the evidence supporting it. Much of what we read about the disinformation potential of LLMs is theoretical, speculative, or anecdotal. The actual experimental work — the patient, systematic testing of what these models really do when prompted to lie — has been surprisingly scarce.

This is precisely the gap that Ivan Vykopal and his colleagues at the Kempelen Institute of Intelligent Technologies in Bratislava set out to fill. Their paper, Disinformation Capabilities of Large Language Models, presented at the 2024 Annual Meeting of the Association for Computational Linguistics, offers one of the most rigorous empirical assessments to date of what the current generation of LLMs can and cannot do as generators of false news — not a manifesto, not a forecast, but a controlled experiment with a clearly defined methodology and reproducible results.

The design is straightforward and, for that reason, compelling. The researchers selected twenty real disinformation narratives drawn from professional fact-checkers — Snopes, Agence France-Presse, the European Digital Media Observatory — spanning COVID-19, the Russo-Ukrainian war, health hoaxes, US elections, and regional narratives. These are not inventions but circulating falsehoods, from the claim that vaccines cause autism to the assertion that the Bucha massacre was staged. The team then prompted ten different language models — including GPT-3, GPT-4, ChatGPT, Llama-2, Mistral, Falcon, and Vicuna — to write news articles supporting each narrative, generating 1,200 texts and subjecting 840 of them to human annotators against a six-question framework measuring coherence, journalistic style, agreement with the narrative, and the generation of novel supporting arguments.

The central finding is sobering. The models are, by and large, perfectly willing and perfectly able to generate convincing disinformation. They produce coherent, well-structured, news-like articles that agree with dangerous falsehoods — and, more disturbingly, they often invent new supporting evidence to do so, hallucinating plausible-sounding names, events, and statistics to lend credibility to the fabrications. This is particularly insidious: it is one thing to repeat a known lie, and quite another to manufacture fresh, fabricated “facts” that a reader would have to independently debunk.

But the most interesting part of the study is where it complicates the simple narrative. The models did not behave uniformly; their willingness to generate disinformation varied dramatically. Some — notably Vicuna and the older GPT-3 Davinci — proved to have essentially no functioning safety filters for this use case, while others showed that safer behavior is achievable: Falcon refused roughly a third of requests and Llama-2 showed a comparatively high refusal rate, with ChatGPT in between. The danger, in other words, is not an inherent and uniform property of the technology; it is a function of how each model was trained and aligned — which means safety is a design choice, not an impossibility. The study also found the models to be steerable through prompt context, and more compliant with regional falsehoods, where less authentic information exists to contradict them. LLMs may thus be especially dangerous for campaigns targeting smaller linguistic communities or fast-moving events, where the protective ballast of well-documented truth is thin.

Yet the paper does not end on a note of unrelieved alarm. Two countervailing observations temper the picture. The generated texts proved quite detectable: the best automated detection models identified machine-generated articles with high precision, suggesting a meaningful layer of defense is technically feasible — at least until adversaries adapt. And, rather elegantly, the researchers showed that the models themselves can be part of the solution, using GPT-4 to partially automate the evaluation of generated texts and pointing toward scalable, repeatable monitoring of model safety.

The honest conclusion resists the pull of both techno-optimism and techno-panic. The capability to generate convincing, dangerous disinformation at scale is real, demonstrated, and present in widely available models — including open-source ones that cannot be recalled or centrally controlled. That is no longer speculation; it is experimental fact. At the same time, the threat is neither uniform nor unmanageable: safety filters work when they are built, generated content remains detectable for now, and the same technology that produces the problem can be enlisted in its mitigation.

Perhaps the most important caveat is the one the authors themselves insist upon: their study is a snapshot, capturing the state of the field at a particular moment with a particular set of models. The technology moves quickly, and the next generation may behave differently. This is the recurring epistemological challenge of the entire domain — we are assessing a moving target, and any honest assessment must carry an expiration date. What Vykopal and his colleagues have given us is not the final word, but something more useful: a rigorous, replicable method for asking the question again as the technology evolves. In a debate too often conducted in the currency of assertion, that methodological contribution may prove as valuable as the findings themselves.