On a late afternoon in Bangui, the capital of the Central African Republic, the air is heavy with heat, dust and the low hum of generators. In the markets, people bargain over cassava and phone credit; on street corners, men argue about politics in French and Sango. It is easy to forget that this weary city has also become a frontline in an invisible war—a war over what people believe. According to a report by Anthropic, the American company behind the artificial intelligence assistant Claude, Russian agents have deliberately used that technology to spread pro-Russian influence in the country. The report, titled “Detecting and Countering Misuse of AI,” was published in September and describes how opposition politicians were monitored, propaganda content was created and disinformation campaigns were supported. In particular, the report points to Radio Lengo Songo, a station that international investigators from All Eyes on Wagner—a collective of journalists, academics and experts who track the Wagner Group and its successor structures, such as the Africa Corps—say is funded by Russia. To an ordinary listener, the station might seem like just another broadcaster playing music, talk and news. But Anthropic says Claude was used to produce content that was overtly pro-Russian, celebrating the government of President Faustin-Archange Touadera and the Russian mercenaries of the Africa Corps, while sharply criticizing France and the opposition. The report does not name the opposition figures who were monitored, nor does it say exactly what data was collected or where it ended up. Yet the story it tells is deeply disturbing. This is not a conspiracy theory whispered in hotels. It is a documented case of artificial intelligence quietly turned into a weapon of perception. For a country that has already suffered so much—coups, rebellions, peacekeepers, mercenaries—this is another reminder that power in the twenty-first century depends on more than bullets and tanks. It depends on stories. And now AI can write those stories in a thousand voices, at almost no cost, without ever growing tired.
Anthropic’s investigation goes deeper than radio scripts. The report says Claude was not merely generating cheerful slogans; it was being used as a tool of surveillance and psychological pressure. Opposition politicians were watched, apparently so the government would have material to use against them later. Radio Lengo Songo and other outlets then used that material to frame those people as enemies of the state, as traitors or foreign agents. Lou Osborn, a member of All Eyes on Wagner who helped research the report, told DW that this fits a pattern seen in the country for years. “There have been several arrests of people who were not necessarily closely affiliated with the Touadera regime,” Osborn said. “They’ve been doing this for several years now.” He described how arrests were followed by online campaigns in newspapers and on the radio, presenting the detainees either as proxies for France or Western intelligence services, or as plotters preparing a coup against Touadera. In the feverish atmosphere of Bangui, such accusations can be as damaging as a prison sentence. The report also alleges that Claude was prompted to describe opposition figures as armed militants, presumably in the hope that the government would feel justified in moving against them. This time, the AI refused. But the fact that someone asked is itself chilling. The attempt to turn a language model into an informer, a propagandist and a false witness is not a freak accident. It is a deliberate method. And the report suggests it worked often enough to be worth repeating. Anthropic says it uncovered these activities and took steps to interrupt them, but the effects on the ground—the fear, the suspicion, the silence—cannot be deleted with a simple account cancellation.
For the opposition in the Central African Republic, the report confirms what many have long believed: that the government in Bangui and its Russian allies are determined to get rid of them. Martin Ziguele, chairman of the Movement for the Liberation of the Central African People, spoke to DW about the climate of fear that now surrounds public life. “We have been observing this since 2020, when the CPC, a coalition of rebel groups, attacked Bangui,” he said. “Strangely enough, several media outlets did everything in their power to establish alleged links between certain opposition leaders and the CPC.” He also recalled campaigns accusing opposition members of being paid by foreign countries, as well as an attempt to pass a law declaring that foreign agents were present in the country. For Ziguele, all of these things belong to the same pattern: a regime that cannot tolerate disagreement, using every available tool—including Russian mercenaries and, now, artificial intelligence—to crush its critics. “The government wants to get rid of us,” he said simply. That sentence carries enormous weight in a country where politicians have been killed, exiled or locked up. It is one thing to be criticized by your rivals. It is another to be systematically destroyed on the radio, in the press and online, with algorithms generating the lies that justify your arrest. The human cost is not abstract. Politicians who might otherwise speak freely now think twice before opening their mouths. Journalists hesitate before publishing a critical story. Ordinary citizens learn that the price of dissent is not just a fine or a harsh remark, but a campaign of public humiliation and the threat of violence. Democracy is not killed in a single blow in places like this. It is strangled slowly, one invented accusation at a time.
Russia, for its part, dismisses the entire report. DW reached out to Dmitri Sytyi, a representative of the Russian mission in the Central African Republic, and he insisted that there was nothing solid in the allegations. “Yes, I heard about the report in the news,” he said. “But there are no concrete facts in it. These are vague allegations, without precision and without evidence.” The denial is predictable. Moscow has spent years presenting Western investigations of its activities in Africa as attempts to smear Russia and steal its influence. The Central African Republic is rich in diamonds, gold and other resources, and Russia has built a close relationship with President Touadera, sending soldiers, weapons and political support. To those familiar with the region, the Russian response may sound like yet another example of the very disinformation the report describes. But the report’s credibility rests on the fact that Anthropic says it found evidence and acted on it, rather than simply making accusations. The company says it uncovered the operations between December 2025 and August 2026, deleted the accounts involved and strengthened its detection systems. Still, in a world of deepfakes, bots and anonymous Telegram channels, almost no piece of evidence can be completely convincing to everyone. The Russian denial may not change anyone’s mind, but it does exactly what denials are meant to do: it sows doubt. In a conflict where information itself is the weapon, the truth becomes the first casualty. The people of Bangui may never know for certain who was behind every anonymous message or every suspicious radio broadcast. But the effect of uncertainty is not paralysis. It is fear. And fear, as any propagandist knows, is the easiest emotion to exploit.
The Central African Republic is not the only country where Anthropic says AI was weaponized. The report describes a broader pattern of AI-assisted influence operations across the African continent, from the Democratic Republic of the Congo to Kenya, Sudan and Mali. In the DRC, researchers identified 318 articles published by a network of around 70 fake news sites, which were then distributed by more than 250 fictitious accounts on X, formerly Twitter. The content supported the Congolese government in its volatile conflict with Rwanda, an attempt to shape both domestic opinion and international perceptions. In Kenya, Claude was allegedly used to generate messages hostile to the opposition in batches of fifty at a time—a kind of digital assembly line for political insults. In Sudan, Anthropic says AI was used in an operation attributed to individuals in the United Arab Emirates, specifically to fabricate witness statements meant for the United Nations. That is not just propaganda; it is an attempt to falsify the historical record. In Mali, an adviser with ties to intelligence services allegedly used Claude to build a surveillance platform designed to monitor telecommunications, targeting 25 million SIM cards while finding a way around the requirement to obtain judicial authorization. Each of these cases shows a different way in which AI can be misused: to flood the media, to silence opposition, to invent evidence, to watch entire populations. None of these operations is enormous in the way that a traditional state propaganda machine used to be. But that is precisely the point. Artificial intelligence makes manipulation cheap, fast and decentralized. A single person with a laptop can now produce the output of an entire troll farm. The report does not claim that Russian agents personally carried out every one of these operations; in some cases, the alleged actors have little connection to Moscow. But taken together, the examples reveal a troubling new reality. AI has become the perfect amplifier for the age-old human instincts of power, greed and cruelty.
Anthropic says it responded by shutting down all of these operations, deleting the accounts involved and reinforcing its detection systems. The company’s willingness to publish its findings is commendable, and its engineers have clearly learned valuable lessons about how easily AI can be abused. But the report also leaves many questions unanswered. How many other operations remain hidden? How effective was Claude’s refusal in those cases where it said no? And what will happen when the next generation of AI becomes even better at mimicking human creativity and tone? If large language models can be used to monitor political opponents, fabricate statements to the United Nations and build surveillance systems without judicial oversight, then the warning signs are impossible to ignore. The people of the Central African Republic have already lived through years of upheaval—from coups to civil war to the arrival of Wagner mercenaries who were supposed to be protectors and became, in the eyes of many, a foreign force of occupation. Now they must wonder whether the lie machine is still working, humming quietly in the background, learning new languages and new ways to deceive. Perhaps the most humanizing lesson of the whole episode is that AI itself is not evil. It is a mirror, reflecting our own desires, fears and weaknesses. When powerful actors use it to distort reality, they do so because it works. The answer is not to abandon the technology, but to hold it accountable—and to hold ourselves accountable. Citizens need media literacy, journalists need support, governments need transparency, and companies like Anthropic need to keep watching. In the end, the fight is not really about algorithms or models. It is about protecting the simple, fragile idea that human beings deserve the truth. In Bangui, in Kinshasa, in Nairobi and far beyond, that idea is still worth defending, even if it means fighting a war of words with machines.