Anthropic published a threat intelligence report on September 10 detailing how its Claude AI models were used across propaganda, political manipulation, and mass surveillance operations touching four African countries: Kenya, the Central African Republic, Mali, and the Democratic Republic of Congo. The report covers malicious activity Anthropic identified and disrupted between December 2025 and August 2026, spanning suspected state-sponsored groups, politically motivated individuals, and commercial actors. The African cases sit inside a much larger global report documenting nine influence-operation cases tracing back to actors in Russia, Iran, Turkey, the Gulf, South Asia, and Europe, targeting audiences across six continents.
Kenya: A Domestic Political Operator, Not a Foreign One
The most detailed African case involved a single domestic actor in Kenya, tracked under the designation GTG-54004, who used Claude to mass-produce political content ahead of the country’s 2027 election. The operator supplied specific political topics, talking points, and hashtags, then had Claude turn that material into large batches of platform-ready posts, generated in sets of exactly 50 tweets designed to mimic organic, grassroots sentiment rather than obviously coordinated messaging. A significant share of the content praised Energy Cabinet Secretary Opiyo Wandayi for halting a scheduled Kenya Power electricity tariff increase, pushed under hashtags including #PowerReliefKE and #PoweringTheNewKenya, while other posts pushed narratives claiming Kenya’s United Opposition coalition was fracturing, directly naming politicians Rigathi Gachagua and former President Uhuru Kenyatta. Anthropic was specific about where AI’s role began and ended here: “The core ideological messages were completely decided by the actors before Claude was used.” In other words, Claude functioned as a production and formatting tool for messaging the operator had already fully scripted, not as the source of the political ideas themselves. Anthropic classified the operation as Category One on its internal Breakout Scale, meaning the activity stayed contained within its own network of fake accounts, and the company said it found no evidence the campaign actually reached or influenced real audiences.
Central African Republic: A Russian State-Linked Operation
The Central African Republic case looked considerably different in both origin and reach. Anthropic removed an account operated by a Russian-speaking actor based in Bangui who the company said provided the production backbone for a Russian state-aligned Foreign Information Manipulation and Interference operation. Anthropic linked the individual to Politology, the Africa Corps and Wagner-linked influence branch assessed to have come under the control of Russia’s Foreign Intelligence Service in late 2023, describing the person as functioning as Politology’s local media coordinator on the ground. The operation ran daily content through Radio Lengo Songo, a local FM station, coordinated alongside Russian state media outlets RT, Sputnik Afrique, and TASS and the Russian House in Bangui, with content amplified further through Telegram channels and picked up by local news outlets. Anthropic classified this operation as Category Four on the Breakout Scale, its most severe rated classification among the African cases, reflecting genuine daily broadcast reach through established local media rather than activity confined to fake accounts.
Democratic Republic of Congo: Commercial Influence for Hire
The DRC case was part of a considerably larger commercial operation spanning six continents rather than a campaign targeting DRC specifically. Anthropic identified and removed an account linked to an operation using Claude to produce and rewrite political content across roughly 70 fabricated news websites, backed by around 70 matching accounts on X and more than 250 inauthentic commenting accounts, publishing at least 8,913 articles across approximately 20 languages. Anthropic traced the operation to LKM Company, a France-based digital advertising agency, describing the activity as an “influence-as-a-service” business model, where private firms can be hired by paying clients to manipulate information, promote political agendas, or run targeted smear campaigns.
What’s Known About the Mali Case
Mali is confirmed as one of the four African countries covered in Anthropic’s report, though detailed specifics of the Malian case weren’t available in the coverage reviewed for this piece. Readers looking for the full breakdown of that case should consult Anthropic’s original report directly.
Why the Nuance Here Actually Matters
It’s worth resisting the urge to flatten these four cases into a single, uniform story, because the actual evidence varies enormously in severity between them. Kenya’s case, the most politically prominent given the country’s approaching 2027 election, was explicitly assessed as having no confirmed real-world reach, activity confined entirely to a network Anthropic itself identified and shut down. The Central African Republic case is a different order of concern entirely, a state-linked operation with genuine daily broadcast reach through established local radio and international coordination with Russian state media. That distinction matters for how seriously each case should be read, and for avoiding the kind of flattening that turns “AI was used to draft this content” into “AI determined this outcome,” which the evidence in at least the Kenyan case doesn’t support.
None of this is unique to Anthropic or Claude either. OpenAI disrupted separate election-related AI misuse targeting Ghana back in January 2025, underscoring that this is an industry-wide challenge tied to how accessible generative AI has become for producing convincing text at scale, not a problem specific to any single AI company’s model.
Why This Kind of Disclosure Matters
Anthropic publishing detailed findings about misuse of its own product, including the specific accounts, tactics, and actors involved, is itself worth noting as a transparency practice, giving researchers, journalists, and election-integrity organisations concrete detail to work from rather than vague warnings about AI’s potential for misuse. With Kenya heading toward a 2027 general election and several other African countries facing their own upcoming votes, reports like this one are likely to keep surfacing as both a genuine early-warning tool and a test of how quickly platforms, regulators, and AI companies can actually act on what they find before contained activity has a chance to reach real audiences at the scale the Central African Republic case already demonstrates is possible.
