Artificial intelligence company Anthropic has disclosed a Kenyan political influence operation in which a single actor used its Claude AI model to produce large batches of social media posts designed to make coordinated political messaging appear as genuine grassroots opinion. In its September 2026 Threat Intelligence Report, Anthropic said it identified and removed an account used by the actor to generate Kenyan political content. The company said the activity was designed specifically to conceal the coordination behind the campaign.

“The operation was built to look like spontaneous, grassroots public sentiment,” Anthropic said.

The disclosure comes as Kenya approaches the 2027 General Election, with political parties and candidates increasingly using social media to communicate with voters and shape public debate.

Anthropic said the actor used Claude across several sessions to generate batches of exactly 50 tweets at a time. The prompts instructed the model to make the posts appear spontaneous and organic rather than part of a coordinated political campaign. The company said the actor provided political themes and talking points, after which Claude was used to produce multiple variations of social media content. This allowed the operator to produce political material at a scale that would have required considerably more time if each post had been written manually. Anthropic described the activity as an example of how generative AI can be used not only to create political content but also to help disguise the coordination behind it.

According to Anthropic, some of the content generated by the operation promoted messages favourable to the Kenyan government. The company said the actor produced posts praising Energy Cabinet Secretary Opiyo Wandayi over the government’s decision to halt a planned increase in electricity tariffs. The campaign also used hashtags including #PowerReliefKE and #PoweringTheNewKenya. Other content targeted politicians associated with the opposition, including former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta, while promoting claims about divisions within the United Opposition. Anthropic said, however, that it had not established the identity of the actor or the organisation behind the operation. It also said it had found no evidence that the Kenyan government was involved. The political character of the posts therefore does not, on its own, establish who commissioned or financed the activity.

Astroturfing is the practice of creating the appearance of spontaneous public support when the messaging is actually organised by a hidden individual or group. In the Kenyan case described by Anthropic, the central feature was the attempt to make centrally generated political messages look like independent opinions from ordinary social media users. That distinction matters because large numbers of similar posts can create the impression that a particular political position has widespread public support even when the messages originate from a relatively small operation. The use of AI makes such activity easier to scale because one set of political talking points can be turned into numerous variations within a short period.

Anthropic’s findings do not indicate that the campaign successfully influenced large numbers of genuine social media users. The company said it detected the activity while it was still developing and removed the account and organisation associated with it. It classified the Kenyan operation at the lowest level of its Breakout Scale, indicating that the activity remained largely contained within the network involved and that Anthropic found no evidence that it had reached or influenced genuine users. That distinction is important. The discovery of an organised influence operation does not necessarily mean that it changed public opinion or affected voters.

The case highlights a developing challenge for political communication as Kenya moves towards the 2027 election. Social media users increasingly encounter political messages through platforms where the identity, motivation and coordination of accounts can be difficult to establish. AI tools can make it easier to produce large quantities of content, adjust language for different audiences and create posts that appear to have been written independently. For news organisations, researchers and social media users, the development increases the importance of examining who is behind an account, whether apparently independent posts share identical talking points, and whether a sudden surge in political content comes from genuine users or coordinated activity. Anthropic’s report does not establish that AI-generated political content is inherently deceptive. The concern in this case was the deliberate use of AI to conceal coordination and create a false impression of grassroots support.

Anthropic’s disclosure should therefore not be interpreted as evidence that any particular Kenyan politician, political party or government agency ordered the operation. The company has identified the activity and described its political messaging, but it has not publicly identified the person behind the account or established who, if anyone, financed or directed the campaign. What is confirmed is that a single actor used Claude to generate batches of Kenyan political posts and instructed the model to make them appear spontaneous. Anthropic said it has removed the account and used the findings to strengthen its systems for detecting similar misuse. The case provides an early indication of how artificial intelligence could be incorporated into political influence campaigns ahead of Kenya’s 2027 election, not necessarily by creating new political messages, but by making coordinated messages faster and easier to produce while attempting to make them look like ordinary public opinion.