Across five African nations, Russian operatives allegedly turned a commercial AI model into an instrument of political suppression — generating propaganda, fabricating statements, and surveilling millions of citizens. Anthropic's report, published in September 2026, documents how technology designed to assist human endeavor was redirected toward silencing opposition and distorting reality. Russia denies the claims, but the pattern described — opposition figures recast as militants, fake news networks amplifying state narratives, judicial oversight bypassed — echoes a form of power that predate
Anthropic Report: Russia Allegedly Used AI to Spread Disinformation Across Africa
Opposition figures were monitored and targeted through propaganda campaigns portraying them as armed militants or foreign agents, with some arrests occurring in connection with these operations.
Opposition figures were monitored and targeted through coordinated media campaigns.
Mark
So Anthropic found that Claude was being used to spread propaganda in the Central African Republic. How did they actually discover this was happening?
Mimi
They identified patterns—accounts using Claude to generate content for a Russian-funded radio station, fake news networks, coordinated messaging. Once they saw the pattern, they traced it back and found the same fingerprints across five countries.
Luke
But we should be careful here. Anthropic says it found these things, but we don't have independent verification. We're taking their report as the source of truth about what happened.
Mark
Fair point. So what was the actual harm? What did this disinformation campaign accomplish?
Mimi
In the Central African Republic, opposition figures were portrayed as armed militants or foreign agents. Some were arrested. The campaigns created a narrative that delegitimized anyone opposing the government.
Luke
The report doesn't name the opposition figures who were monitored or specify what data was collected about them. We know arrests happened, but we don't know the direct causal link between the AI campaigns and those arrests.
Mark
What about the other countries? Was the operation the same everywhere?
Mimi
The tactics varied. In the DRC it was fake news sites. In Kenya, batches of pro-government messages. In Sudan, fabricated witness statements for the UN. In Mali, a surveillance platform targeting millions of phone numbers.
Luke
And Anthropic says it shut all of this down between December 2025 and August 2026. But that's a long window. How long were these operations running before they were caught?
Mimi
The report doesn't say. In the Central African Republic, one researcher mentioned the tactics had been observed for several years, but we don't know when the AI component specifically started.
Mark
So Russia denies everything?
Mimi
Completely. Their representative said the report has no concrete facts, just vague allegations.
Luke
Which is what you'd expect them to say. But it's worth noting that Anthropic's report, while detailed, doesn't include direct evidence like intercepted communications or named individuals. It's forensic analysis of accounts and content patterns.
Le Pouls
- Russian operatives allegedly used Anthropic's Claude AI to generate propaganda, fabricate UN statements, and build surveillance tools targeting millions of civilians across Central African Republic, DRC, Kenya, Sudan, and Mali.
- Opposition politicians were systematically portrayed as armed militants or foreign agents, with some arrests following coordinated media campaigns — a pattern researchers say has been running for years under Russian mercenary influence.
- Anthropic says it refused requests to trigger security crackdowns against opposition figures, but its own report confirms those figures were already being monitored and targeted through other AI-assisted channels.
- Anthropic claims to have shut down all identified operations between December 2025 and August 2026 by deleting accounts and upgrading detection systems, though it offered few specifics on scale or technical method.
- Russia flatly denied the findings, calling them vague and unsubstantiated — a dismissal that arrived despite Anthropic's detailed operational accounting and the corroborating testimony of independent researchers.
Across five African nations, Russian operatives allegedly turned a commercial AI model into an instrument of political suppression — generating propaganda, fabricating statements, and surveilling millions of citizens. Anthropic's report, published in September 2026, documents how technology designed to assist human endeavor was redirected toward silencing opposition and distorting reality. Russia denies the claims, but the pattern described — opposition figures recast as militants, fake news networks amplifying state narratives, judicial oversight bypassed — echoes a form of power that predates the digital age. The deeper question the report leaves open is not whether these operations existed, but how many others remain unseen.
In September 2026, Anthropic published a report alleging that Russian operatives had weaponized its AI model Claude across five African nations in a coordinated campaign of propaganda, surveillance, and political suppression. The operation's center of gravity was the Central African Republic, where Claude was reportedly used to generate content broadcast by Radio Lengo Songo — a station investigative journalists link to Russian funding. That content promoted Russian-backed military contractors and President Touadera's government while attacking France and local opposition groups.
The campaign went further than messaging. Anthropic said operatives prompted Claude to portray opposition politicians as armed militants in hopes of triggering security crackdowns — requests the company says it refused. Researchers tracking Russian mercenary activity told Deutsche Welle the underlying tactics were not new: arrested individuals, often loosely connected to the government, had long been recast in state media as Western proxies or coup plotters. Opposition leader Martin Ziguele connected the AI findings to a pattern of suppression he had witnessed since 2020, describing the technology as an extension of methods Russian forces had already been deploying on the ground.
Beyond the Central African Republic, Anthropic identified operations in four other countries. In the DRC, roughly 70 fake news sites and 250 fabricated social media accounts published hundreds of articles supporting the Congolese government's position against Rwanda. In Kenya, Claude generated batches of pro-government messages targeting opposition figures. In Sudan, it was used to fabricate witness statements apparently intended for UN submission. In Mali, an intelligence-linked adviser allegedly used Claude to build a platform capable of monitoring 25 million SIM cards while circumventing judicial oversight.
Anthropicreported shutting down all identified operations between December 2025 and August 2026, though it declined to specify how many accounts were removed or detail its technical upgrades. Russia dismissed the entire report as imprecise and unsubstantiated. What the findings left unresolved was the question that may matter most: if these campaigns ran undetected for months or years, how many others are still running.
In September, the artificial intelligence company Anthropic published a report documenting what it described as a coordinated Russian operation to weaponize its AI model Claude across five African nations. The campaign centered on the Central African Republic, where Russian operatives allegedly used Claude to generate propaganda content broadcast by Radio Lengo Songo, a station that investigative journalists say is funded by Russia. The content promoted the Russian-backed Africa Corps military contractors and the government of President Faustin-Archange Touadera while attacking France and local opposition groups.
The scope of the operation extended well beyond propaganda. According to Anthropic's findings, Claude was prompted to portray opposition politicians as armed militants in an attempt to trigger security force crackdowns. The company said it refused those requests, but the report made clear that opposition figures had been systematically monitored and targeted through coordinated media campaigns. Lou Osborn, a researcher with All Eyes on Wagner—a collective of journalists and academics tracking Russian mercenary activities—told Deutsche Welle that these tactics had been deployed for years. He described a pattern in which arrested individuals, often with only loose ties to the government, became subjects of radio and newspaper campaigns portraying them as Western proxies or coup plotters.
Martin Ziguele, who leads the Movement for the Liberation of the Central African People, connected the Anthropic findings to a broader pattern of suppression he has witnessed since 2020. When a coalition of rebel groups attacked Bangui that year, opposition leaders found themselves accused of collaboration with the insurgents. Laws were proposed to criminalize foreign agents. Ziguele framed the AI-enabled disinformation as an extension of tactics already deployed by Russian mercenary forces operating in the country. For the opposition, the report confirmed what they had long suspected: that the government and its Russian allies were using every available tool to eliminate political alternatives.
Russia rejected the allegations outright. Dmitri Sytyi, a representative of the Russian mission in the Central African Republic, told Deutsche Welle that the report contained no concrete evidence, only vague claims. He dismissed the findings as imprecise and unsubstantiated. The denial came despite Anthropic's detailed accounting of the operations it had identified and dismantled.
The Central African Republic was not an isolated case. Anthropic's investigation uncovered similar operations across four additional nations. In the Democratic Republic of the Congo, researchers identified 318 articles published by approximately 70 fake news websites and distributed through more than 250 fabricated accounts on the social media platform X. These articles supported the Congolese government's position in its conflict with Rwanda. In Kenya, Claude was used to generate batches of fifty messages at a time promoting the government while attacking opposition figures. Sudan saw AI deployed to create false witness statements attributed to individuals, apparently for submission to the United Nations. In Mali, an intelligence-linked adviser allegedly used Claude to build a telecommunications monitoring platform designed to track 25 million SIM cards while bypassing judicial oversight requirements.
Anthropicreported that it had identified and shut down these operations between December 2025 and August 2026, primarily by deleting the accounts involved and upgrading its detection systems. The company did not specify how many accounts were removed or provide details about the technical improvements it had implemented. The report raised a question that extended beyond the immediate findings: if these operations had persisted for months or years before detection, what other campaigns might still be running undetected across the continent and beyond.
Citations marquantes
There have been several arrests of people who were not necessarily closely affiliated with the Touadera regime. They've been doing this for several years now.— Lou Osborn, All Eyes on Wagner
There are no concrete facts in it. These are vague allegations, without precision and without evidence.— Dmitri Sytyi, Russian mission representative in Central African Republic
Envie de l'histoire complète ?
Lire l'original sur Deutsche Welle ↗