A new threat intelligence report by artificial intelligence company Anthropic has exposed a coordinated political influence operation in Kenya that used Claude to mass-produce social media posts designed to look like genuine grassroots opinions.

This comes at a time the government has seriously lost the war on information. The state is now using artificial intelligence to manufacture the appearance of public support for the President and other government alied political figures and narratives ahead of the 2027 General Election.

In its September 2026 threat intelligence report, Anthropic said it identified and removed an account operated by a single actor who was using Claude across several sessions to generate Kenyan political content in batches of exactly 50 tweets. The actor instructed the AI model to make the posts appear as spontaneous grassroots commentary rather than material generated as part of a coordinated political campaign.

The operation, which Anthropic designated GTG-54004, focused on some of the most politically sensitive issues in Kenya, including the cost of electricity, the performance of the government and the state of opposition politics ahead of the 2027 elections. Anthropic said a large portion of the content praised Energy and Petroleum Cabinet Secretary Opiyo Wandayi after the government withdrew Kenya Power’s proposed electricity tariff review.

The timing and subject matter are significant because the government did in fact withdraw Kenya Power’s tariff review application in June 2026, with Wandayi announcing that the current electricity tariffs would remain in force after consultations within government and with stakeholders. The Energy Ministry said the decision was aimed at protecting households, businesses and industries from increased electricity costs.

But according to Anthropic, the problem was not the government's actual decision to withdraw the tariff proposal. It was the attempt to manufacture the appearance of ordinary Kenyans independently celebrating that decision online.

Anthropic said the same network also pushed narratives claiming that Kenya’s United Opposition was breaking apart ahead of the 2027 election, with the messaging directly targeting former President Uhuru Kenyatta and former Deputy President Rigathi Gachagua. The company said the operation used a single playbook to push both pro government and anti opposition narratives, a combination it described as consistent with a coordinated electoral communications effort.

According to Anthropic, the person behind the campaign supplied the political topics, talking points and hashtags before using Claude to turn them into 50 themed, character-counted posts that could then be published across social media. In several sessions, the material was packaged into an interactive copy all format, effectively preparing large quantities of content for deployment.

In other words, Anthropic says the AI was not deciding the political agenda. The human operators decided what they wanted Kenyans to talk about and then used Claude to produce the volume of material required to make those talking points appear as if they were coming from many ordinary people.

Anthropic said this showed that the principal value of Claude to the operation was volume and the appearance of authenticity, with the ideological messages already determined by the actors before the AI was used.

That detail changes the way the operation should be understood. It was not simply a case of someone asking an AI chatbot to write a political tweet. Anthropic describes a system in which pre determined political messages were repeatedly transformed into dozens of apparently independent posts, allowing one actor to produce content at a scale that would otherwise require a much larger team.

The company said the campaign was also connected to an unusual commercial side operation. The same AI workflow was used to market Kenyan retail brands under the personas “SHANKI” and “Elkins Marketer.” Anthropic said the template was used verbatim for retail marketing, including a promotional broadcast that was repackaged as supposedly organic tweets, with links inserted into every fifth post.

That finding suggests the operators may have been applying the same social media production system to both commercial marketing and political messaging, turning a technique normally associated with advertising into a tool for manufacturing political commentary.

Anthropic said the operation also fits a wider pattern it has observed in Kenya in which agencies pay local influencers to manipulate narratives. The company linked this observation to reporting on organised attempts to shape online conversations through coordinated influencer activity.

The Kenyan campaign was assessed using Anthropic's Breakout Scale and classified as Category One, the lowest category in the company's assessment of the real world impact of influence operations. Anthropic said the activity remained confined to a network of fake accounts and local influences on a single platform and failed to reach or influence any real people.

That finding is just as important as the exposure itself.

The report does not say that the operation successfully changed public opinion, influenced an election or persuaded genuine Kenyan voters. Instead, Anthropic says it failed to break out of its artificial network and did not reach real people in a meaningful way.

Anthropic said it discovered the Kenyan campaign after receiving a tip from OpenAI concerning what it described as recidivist activity on OpenAI's platform. The tip prompted an internal investigation into suspected coordinated inauthentic behaviour in Kenya, leading Anthropic to identify the operation.

The company subsequently removed the account and the organisation behind it and said it had built new detection systems around the behaviour observed in the campaign.

The case now presents a difficult question for Kenya as the country moves closer to the 2027 General Election.

If a single actor can take an already prepared political message, feed it into an AI system and obtain batches of 50 apparently human written posts designed to look like spontaneous public opinion, how many political conversations taking place online should Kenyans assume are genuine?

The Anthropic report exposes a method that can make coordinated political messaging look like ordinary citizens speaking independently, particularly when the content is spread through accounts that appear unrelated.

The case also puts greater attention on the growing role of AI in Kenya's political communication industry, where the distinction between genuine grassroots mobilisation, paid influence, coordinated online campaigns and artificially generated commentary may become increasingly difficult for ordinary users to identify.