Artificial intelligence safety research company Anthropic has disrupted a Kenyan political propaganda operation that allegedly used its Claude AI model to generate pro-government messaging and attack opposition figures. The company disclosed the operation in a detailed threat intelligence report examining sophisticated attempts to misuse artificial intelligence for activities including cyberattacks, coordinated influence campaigns, surveillance, biological research and weapons development. Among the operations identified was a Kenya-based campaign that Anthropic described as coordinated, deceptive and manipulative.
Kenyan operator used Claude to generate political content
According to Anthropic, the operation was linked to a single domestic actor who used Claude across multiple sessions to produce large volumes of political content designed to appear as genuine public opinion. The campaign, identified as GTG-54004, involved generating groups of 50 tweets at a time. The operator allegedly instructed Claude to make the posts appear as spontaneous grassroots commentary rather than content produced as part of an organised political campaign.
The actor reportedly supplied political topics, hashtags and specific talking points before using the AI model to develop content that could be published across social media platforms. Anthropic said it identified and removed the account involved in the operation and subsequently developed additional detection systems based on the campaign’s behaviour.
Campaign praised Opiyo Wandayi
A significant part of the campaign focused on Kenya’s political affairs, including Energy and Petroleum Cabinet Secretary Opiyo Wandayi. Anthropic said the operation generated numerous posts praising Wandayi following his decision to stop a planned electricity tariff review by Kenya Power.
The campaign allegedly used hashtags including #PowerReliefKE and #PoweringTheNewKenya to increase the visibility of the messaging. The operation also promoted narratives suggesting that Kenya’s United Opposition coalition was experiencing divisions ahead of the 2027 General Election. Former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta were among the opposition-linked political figures targeted by the campaign.
No evidence of direct government involvement
Despite the campaign’s apparent pro-government position, Anthropic said it had not found evidence linking the operation directly to the Kenyan government. The company classified the activity as a domestic political astroturfing operation, noting that the messaging and hashtags appeared to align with the ruling coalition but that investigators could not establish the exact organisation behind it. Anthropic said the individuals responsible for the operation had not been conclusively identified.
AI workflow also used for commercial marketing
Investigators also discovered that the operator appeared to use the same AI-assisted workflow for commercial marketing campaigns involving Kenyan retail brands. The online activity was reportedly associated with personas including “SHANKI” and “Elkins Marketer.” Anthropic said the discovery helped investigators establish similarities between the political messaging operation and other activities conducted by the same actor. The investigation was also informed by a tip from OpenAI concerning suspected repeat misuse on its platform, according to Anthropic.
Campaign had limited real-world impact
Despite the scale and structured nature of the operation, Anthropic said the campaign did not achieve significant real-world influence. The company assessed the operation using its Breakout Scale and classified it as Category One. According to the report, the synthetic content remained largely contained within a network of fabricated accounts and local influencers on a single platform and did not gain meaningful traction among genuine voters.
Anthropic strengthens Claude’s safeguards
Anthropic said lessons from the investigation have been incorporated into Claude’s safety and detection systems to help prevent similar misuse in the future. The company acknowledged that highly sophisticated misuse cases remain relatively uncommon compared with ordinary activity on its platform. However, it said such incidents demonstrate the emerging ways advanced AI systems can be exploited to manufacture political narratives and simulate public support.
Anthropic said publishing details of the investigation was intended to help other technology platforms identify similar campaigns while providing greater public awareness of emerging AI-related security threats. The Kenyan case also highlights growing concerns over the use of generative AI in political communication, particularly as countries prepare for major elections and political actors increasingly turn to digital platforms to shape public opinion.
The development comes as Kenya continues to explore the use and regulation of artificial intelligence, including efforts to strengthen cybersecurity and develop national frameworks for emerging technologies.
