Anthropic’s latest threat intelligence report has detailed how its Claude AI models were used in propaganda, political manipulation and mass surveillance operations involving four African countries.
The cases, involving the Central African Republic (CAR), Kenya, Mali and the Democratic Republic of Congo (DRC), show how AI was integrated into activities ranging from Russian-backed information operations and political astroturfing to the development of a surveillance platform designed to monitor about 25 million SIM cards.
Published on September 10, the report covers malicious activity that Anthropic identified and disrupted between December 2025 and August 2026. It includes cases involving suspected state-sponsored groups, politically motivated individuals and commercial actors.
AI-powered propaganda in Central African Republic
In the Central African Republic, Anthropic removed an account operated by a Russian-speaking actor in Bangui who it said provided the production backbone for a Russian state-aligned foreign information manipulation and interference operation.
The actor ran a daily content operation through Radio Lengo Songo, an FM station broadcasting on 98.9 FM, while coordinating with Russian state media outlets RT, Sputnik Afrique and TASS, as well as the Russian House in Bangui.
Claude was used to generate stories containing pro-Russia, anti-France, pro-CAR government and pro-Wagner narratives, while the actor instructed the model to make the content appear like ordinary journalism rather than AI-generated material.
Anthropic said its investigation linked the actor to Politology, the Africa Corps/Wagner influence branch that it assessed came under the control of Russia’s Foreign Intelligence Service in late 2023.
The operation also used Claude for internal management, including generating employment contracts, job descriptions, staff scoring systems and recommendations on which employees should be retained or dismissed.
The actor used the model for other political activities, including tracking CAR opposition figures, preparing talking points for Russian House officials and producing forged CAR government documents, including communications attributed to the Gendarmerie and Ministry of Defence.
Claude refused one of the operation’s more aggressive requests to identify real people as militants in a way that could trigger security action. The actor instead changed the request to use anonymous sources.
Anthropic classified the CAR operation as Category Four on the Brookings Institution’s Breakout Scale because the content was broadcast daily through Radio Lengo Songo and amplified through Telegram channels and local news outlets.
Kenya’s AI-generated political astroturfing
In Kenya, Anthropic identified and removed an account operated by a single actor who used Claude to mass-produce political content designed to look like spontaneous grassroots opinion.
The actor generated batches of exactly 50 tweets, instructing Claude to make them appear spontaneous rather than part of a coordinated campaign.
Much of the content praised Energy Cabinet Secretary Opiyo Wandayi over a scheduled Kenya Power electricity tariff increase, using the hashtags #PowerReliefKE and #PoweringTheNewKenya.
The operation also promoted claims that Kenya’s United Opposition coalition was breaking apart ahead of the 2027 general election, targeting politicians including former deputy president Rigathi Gachagua and former president Uhuru Kenyatta.
Anthropic said it found no evidence of government involvement and assessed the activity as an entirely domestic Kenyan operation. However, its pro-administration tone and use of specific hashtags suggested it was plausibly aligned with the ruling coalition, although the exact organisation behind it was not identified.
The same AI workflow was also used to produce marketing content for Kenyan retail brands.
Read more: Kenyan authorities contain cyberattack that hit multiple government websites
According to Anthropic, the political actors supplied the topics, talking points and hashtags, while Claude’s main contribution was turning them into large batches of polished posts that created the appearance of authentic public opinion.
The operation was classified as Category One because it remained confined to fake accounts and local influences on a single platform, with no evidence that it reached or influenced real people.
Mali surveillance platform targeted 25 million SIM cards
The most extensive African case detailed in the report involved Mali, where Anthropic said Claude was used as the primary engineering workforce for a national domestic surveillance platform built for the country’s state intelligence service.
A single Claude subscriber, likely a Bamako-based independent consultant working with Mali’s Agence Nationale de la Sécurité d’État (ANSE), used Claude to build a system called Lakana 360.
Anthropic said the platform was designed to monitor roughly 25 million SIM cards across all three of Mali’s national mobile operators.
The system was also designed to circumvent legal restrictions requiring a court order for disclosure of certain surveillance records. Claude was used to generate intelligence dossiers on tasked phone numbers without requiring ANSE users to provide evidence of valid legal authorisation.
The platform was designed to collect call records, SMS and voice traffic, while also supporting capabilities such as identifying voices across SIM cards, flagging users of encryption and VPN services, identifying potential clandestine meetings and creating geofenced watch lists.
It could also match people against Mali’s biometric civil registry and other government databases.
Anthropic said the surveillance platform itself operated on-premises using local models, meaning Claude was used to help design and engineer the system rather than serving as the model deployed for the surveillance operation.
The company banned the Claude account involved, although the action did not take the already-deployed surveillance platform offline.
The report also cited documentation by the US State Department and Human Rights Watch concerning the detention and abduction of opposition figures, journalists and civil-society members by Malian security services.
AI-generated fake news network targets DRC
The fourth African case involved the Democratic Republic of Congo as part of a wider commercial influence operation spanning six continents.
Read also: Airtel launches Africa’s first satellite-to-mobile service in DR Congo
Anthropic identified and removed an account linked to an operation that used Claude to produce and rewrite political content across about 70 fabricated news websites.
The network also operated about 70 matching X accounts and more than 250 inauthentic commenting accounts. It published at least 8,913 articles in about 20 languages.
Anthropic traced the operation to LKM Company, a France-based digital advertising agency, and described the activity as an “influence-as-a-service” operation in which private companies could be hired to manipulate information, promote political agendas or conduct targeted smear campaigns.
Claude was used to write original articles and rewrite legitimate journalism into politically slanted versions tailored to particular audiences.
The network repeatedly used the same source material in opposing ideological directions, added political angles to stories that originally had none and republished stories across borders after stripping away their original context.
The websites used fabricated journalist names and AI-generated profile photos to make the outlets and social media accounts appear authentic.
The DRC was one of the operation’s four specific target countries, alongside the United States, Brazil and France.
Within the DRC, Anthropic identified 318 articles that generally supported the DRC government’s position and focused on regional mineral deals and tensions with Rwanda.
On September 11, 2025, the network published almost identical articles about the DRC-Rwanda conflict across multiple websites within three minutes, changing the tone for different regional audiences and coordinating their distribution across X.
Anthropic said there were indications that the activity might reflect the interests of one or more customers with a stake in the DRC-Rwanda conflict. However, it could not independently confirm who commissioned the content and found no evidence that any government directed the operation.
The campaign was classified as Category Two because its content remained within the network’s own websites and associated social media accounts, with no evidence of wider breakout to an authentic audience.
Anthropic Reacts
Across the four cases, Anthropic said the operations demonstrate how AI can increase the speed and scale of activities that previously required significantly more human labour.
However, the company also noted that many influence operations it detected failed to reach genuine audiences. In several cases, Anthropic disrupted the campaigns before they could build an authentic following, while the widest reach occurred where state media and other established distribution channels were involved.
Anthropic said it removed the accounts involved in the African operations and used information gathered from the investigations to strengthen its systems for detecting similar misuse.
Technext Newsletter
Get the best of Africa’s daily tech to your inbox – first thing every morning.
Join the community now!
