Anthropic documented the use of its artificial intelligence model Claude to develop a surveillance system in Mali and carry out targeted political content operations in several African countries, in addition to a campaign linked to the war in Sudan. The company said its investigations covered activity it detected and disrupted between December 2025 and August 2026, while some campaigns had limited reach and were not shown to have reached a real audience.
Lakana 360 communications surveillance platform in Mali
In Mali, Anthropic said it was likely that a single subscriber, working as an independent consultant in Bamako for the National Security Agency, used Claude as an engineering tool to build the Lakana 360 platform. According to the company’s report, the platform was designed to monitor data from about 25 million SIM cards across the country’s three telecom operators.
The company said the system was designed to bypass the requirement for a court order to obtain certain surveillance data. It could prepare intelligence profiles for specific numbers and identify voices through SIM cards. Its functions also included monitoring users of encryption services and virtual private networks, creating geographic watchlists and linking data to the biometric civil registry.
Disabling the Claude account did not stop the local platform
It said banning the Claude account disrupted the platform’s design and development, but did not stop the product deployed locally because it operated using local models outside its platform. In the Central African Republic, Anthropic detected an operation in Bangui run by a Russian-speaking individual who produced daily content for Lengo Songo radio in coordination with RT, Sputnik Afrique, TASS and the Russian House.
The company said the operator requested messages supporting the government and Russia and opposing France and the opposition, while removing features that could reveal the texts had been generated by artificial intelligence. According to the report, the use extended to managing staff, preparing employment contracts and job descriptions, evaluating articles, tracking opposition figures, preparing talking points for officials and forging government documents.
Anthropic classified the operation at level four on its influence-operations reach scale, citing the daily broadcasting of content and its amplification through Telegram channels and local media.
Network of fake websites produces political propaganda in the Congo
In the Democratic Republic of the Congo, the company said a commercial network of about 70 fake news websites and accounts on X used Claude to write articles and rewrite real news with different political leanings, attributing them to fictitious journalists.
Political post campaign targets Kenya’s upcoming elections
In Kenya, the company detected an account that used the model to produce batches of 50 political posts each, written to appear to be spontaneous public opinions. The messages praised Energy and Petroleum Minister Opiyo Wandayi, promoted divisions within the United Opposition coalition ahead of the 2027 elections, and targeted politicians including former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta.
Anthropic said it found no evidence of Kenyan government involvement and assessed that the operation was domestic, although its tone may have aligned with the governing coalition. It classified the campaign at level one because it remained confined to fake accounts, with no evidence that it reached or influenced real users.
In an operation linked to the war in Sudan, the company said it had linked an operating entity with a high degree of confidence to officials in an Arab country. According to the report, the entity operated a network of about 300 fake influential accounts and impersonated a real Sudanese rights organization. It added that Claude was used to write testimonies to the United Nations in the names of two people, with the aim of concealing the entity’s connection to one side of the conflict and presenting the testimonies as independent.
The company did not confirm that the testimonies or targeting dossiers reached their intended recipients. Anthropic concluded that the detected cases showed artificial intelligence’s ability to reduce the human cost and accelerate the production of propaganda and surveillance tools. At the same time, it noted that disabling accounts does not remove tools developed locally and deployed outside its platform.
