Claude AI

Claude AI linked to a political influence campaign in Kenya

 Claude AI was used to generate social media content as part of a coordinated political influence operation targeting Kenyan online conversations. This is according to an investigation into the use of the AI model by Anthropic.

The operation used Claude to turn supplied political talking points, topics and hashtags into large volumes of posts designed for publication across social media. Investigators found that the actor provided the AI with the subject matter and instructions before using Claude to produce 50 themed, character-counted posts. The posts were formatted for cross-platform publishing, with some sessions packaging the content into an interactive copy-all widget ready for deployment.

The findings offer a glimpse into how generative AI tools can be incorporated into organised influence campaigns in Kenya, where political debate increasingly takes place on social media.

Claude used to produce political messaging

According to the investigation, the network concentrated its messaging on Kenyan political issues and prominent political figures.

A significant portion of the AI-generated posts praised Energy Cabinet Secretary Opiyo Wandayi over the withdrawal of a proposed electricity tariff increase. The campaign used hashtags including #PowerReliefKE and #PoweringTheNewKenya to increase the visibility of the messaging.

The same network also circulated claims that Kenya’s United Opposition coalition was falling apart ahead of the 2027 General Election. They targeted former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta.

The use of Claude appears to have been focused on content production rather than autonomous decision-making. The actor supplied the subject, talking points and hashtags, while the AI was used to turn those instructions into multiple variations that could be deployed through the network.

Identical messages pushed through fake accounts

The investigation found evidence of a highly structured political astroturfing campaign in which identical or closely coordinated messages were distributed across different conversations.

The network consisted of fake accounts and local influencers operating on a single platform. Investigators classified the operation as Category One on the Breakout Scale. They noted that the activity remained isolated within the network and failed to reach or influence real people.

Some of the coordinated activity focused on amplifying posts concerning Wandayi and the electricity tariff issue. Screenshots included in the investigation show several accounts posting similar messages around the same news developments, with some accounts amplifying content from established Kenyan media pages.

The coordinated accounts appeared to be pushing a common narrative in support of the government’s position on electricity prices.

Same AI workflow used for Kenyan brands

The actor reportedly used an almost identical AI workflow to create content for Kenyan retail brands. They were operating under a marketing persona identified as “SHANKI” or “Elkins Marketer.”

This suggests that the use of Claude was not limited to political messaging. The same underlying process could be used to rapidly generate marketing content, with the political operation applying the workflow to political talking points, hashtags and narratives.

However, investigators said they found no evidence of government involvement in the broader activity. They assessed that the operation appeared consistent with an entirely domestic Kenyan operation.

Who was behind the operation?

The investigation did not establish the real-world identities of the people responsible. Investigators said the network’s pro-administration tone and use of particular hashtags suggested that it was plausibly aligned with Kenya’s ruling coalition. However, they did not identify the organization responsible for the campaign.

This is an important qualification because the findings establish the nature of the activity without conclusively attributing it to a particular political party, government agency or organization.

The operation was ultimately assessed to have had limited impact. Despite the coordinated use of fake accounts and AI-generated material, investigators found that the activity did not extend beyond the network sufficiently to influence genuine users.

What the Claude case means for Kenya

Generative AI makes it considerably easier to produce large quantities of reasonably consistent social media content. A campaign that previously required multiple people to draft posts, rewrite messages and adapt content to different formats can now use an AI model to generate dozens of variations from a common set of instructions.

In this case, Claude was used to produce 50 themed posts from supplied talking points and hashtags, demonstrating how AI can fit into an existing influence operation without being the operation itself.

The limited reach of this particular campaign also shows that generating large volumes of content does not automatically translate into influence. Fake accounts still need to attract genuine audiences, gain credibility and move conversations beyond the coordinated network.

For Kenya, however, the development raises questions about how political campaigns, marketers and online influence actors will use generative AI as the country moves closer to the 2027 General Election.