Anthropic Reveals AI-Driven Political Influence Operation In Kenya

An artificial intelligence-driven political influence operation in Kenya used Anthropic’s Claude to generate batches of social media posts supporting the government and attacking opposition figures, according to a new report by Anthropic.

The company said it identified and removed an account operated by a single actor that was used to mass-produce Kenyan political content. The campaign was designed to make coordinated messaging appear to be spontaneous public sentiment.

The actor used Claude across multiple sessions to generate batches of exactly 50 tweets, with instructions to make them resemble organic grassroots commentary rather than coordinated political communications.

A significant portion of the content praised Energy Cabinet Secretary Opiyo Wandayi over a planned Kenya Power electricity tariff increase. The posts used hashtags including #PowerReliefKE and #PoweringTheNewKenya to increase visibility.

The operation also promoted claims that Kenya’s United Opposition coalition was falling apart ahead of the 2027 General Election, while targeting politicians Rigathi Gachagua and former President Uhuru Kenyatta.

Anthropic said the same AI workflow was separately used by the actor for marketing Kenyan retail brands under the personas “SHANKI” and “Elkins Marketer.” The company found no evidence of government involvement and said the activity appeared to be a domestic Kenyan operation.

The company classified the campaign as Category One on the Breakout Scale, meaning the activity remained confined to the network of fake accounts and local influencers on a single platform and did not reach or influence real people beyond that network.

Anthropic said it could not establish the identities of those behind the campaign. However, the operation appeared to be aligned with the ruling coalition based on its pro-administration tone and use of specific hashtags, although the company said it had not identified the organisation responsible.

The investigation found that the campaign used one playbook to amplify both pro-government narratives and messages undermining the opposition. Anthropic said this was consistent with a coordinated electoral communications effort.

The AI model was not responsible for creating the campaign’s political objectives. According to Anthropic, the operators supplied the topics, talking points and hashtags, then used Claude to turn them into 50 themed and character-counted posts ready for publication.

This distinction is significant because it shows how generative AI can amplify an existing political strategy rather than independently determining the message. Anthropic said the campaign’s ideological positions had already been decided by the actors before Claude was used.

The workflow also had commercial applications. Anthropic found that a template used for political messaging was reused for retail marketing, including promotional content repackaged to appear as organic tweets, with links inserted into every fifth post.

Anthropic said the operation was discovered following a tip from OpenAI concerning repeat activity on its platform. The company subsequently investigated the suspected coordinated inauthentic behaviour, removed the account and organisation behind it, and developed detection systems based on the campaign’s behavioural signature.

The Kenyan case forms part of Anthropic’s broader September 2026 report on the misuse of Claude by political influence networks, cybercriminals and state-linked actors.

The findings point to a growing challenge for Kenya ahead of the 2027 election cycle. AI can significantly lower the cost and time required to produce large volumes of politically targeted content, making it harder for audiences and platforms to distinguish genuine public opinion from coordinated campaigns.

For Kenya’s political and technology ecosystem, the episode also highlights a wider question: as AI makes political messaging cheaper and more scalable, how will platforms, regulators, political parties and voters identify coordinated influence before it shapes public debate?