Anthropic: How Claude was used in Kenya to create fake political support

An artificial intelligence-powered operation in Kenya attempted to manufacture the appearance of grassroots political support ahead of the country's 2027 General Election, according to a new threat intelligence report by AI company Anthropic.

The operation involved the use of Anthropic's Claude AI to generate large batches of political posts designed to appear as though they had been written independently by ordinary Kenyans.

Anthropic said it identified and disrupted the campaign, removing the account used in the operation and the organisation behind it.

The company said, however, that the campaign had very limited impact.

Its assessment found that the activity remained isolated within a network of fake accounts and local influences on a single platform and failed to reach or influence real people.

The findings form part of Anthropic's most detailed threat intelligence report to date, which examines how its AI systems have been misused for activities including cyberattacks, influence operations, surveillance, biological misuse and weapons-related activity.

Anthropic said it disrupted every operation highlighted in the report and used lessons from the investigations to strengthen its safeguards.

According to Anthropic, the Kenyan operation was conducted by a single actor who used Claude to mass-produce political content.

The actor would provide the AI with the subjects, talking points and hashtags before instructing it to turn them into batches of exactly 50 posts.

The instructions were specifically designed to make the content appear like spontaneous grassroots commentary rather than part of a coordinated political campaign.

Anthropic classified the activity as a domestic Kenyan coordinated inauthentic behaviour campaign and designated it GTG-54004.

A significant portion of the content praised Energy Cabinet Secretary Opiyo Wandayi, particularly over the government's handling of a scheduled Kenya Power electricity tariff increase.

The campaign used hashtags including #PowerReliefKE and #PoweringTheNewKenya to amplify the messaging.

The operation also pushed narratives suggesting that Kenya's United Opposition coalition was falling apart ahead of the 2027 election, with politicians including Rigathi Gachagua and former President Uhuru Kenyatta among those targeted.

Anthropic said the same AI-assisted workflow was separately used to promote Kenyan retail brands under marketing personas identified as “SHANKI” and “Elkins Marketer.”

This overlap was significant because it showed that the operator was not using Claude exclusively for political messaging.

The same basic system was used to turn promotional material into posts designed to look like organic public commentary.

Anthropic said its investigation found no evidence that the Kenyan government was involved in the operation.

The company described the activity as appearing to be an entirely domestic Kenyan operation.

It said the pro-administration tone of the content and the hashtags used suggested that the campaign was plausibly aligned with the ruling coalition, but investigators had not identified the specific organisation responsible.

That distinction means the report does not establish that the government, the ruling coalition or any particular political party commissioned the operation.

Instead, Anthropic's assessment is that the operation was a local Kenyan political astroturfing campaign whose messaging favoured the administration while undermining the opposition.

One of the most revealing findings is that Claude did not appear to be deciding the campaign's political messages.

The people behind the operation supplied the ideological positions, topics and talking points themselves.

Claude's role was primarily to increase the volume and appearance of authenticity.

Anthropic said the actor repeatedly used the model to humanise and refine batches of 50 pre-drafted topics and posts, including formatting them according to character limits and preparing them for publication.

In some cases, the resulting material was packaged into an interactive copy-all format that could be used for deployment.

This illustrates one of the emerging concerns around generative AI in elections: the technology does not necessarily need to determine what people believe.

It can instead make it substantially easier to produce large amounts of seemingly authentic content supporting a predetermined political message.

Anthropic said its investigation was triggered partly by information shared by OpenAI concerning related activity on its platform.

The company subsequently investigated suspected coordinated inauthentic behaviour in Kenya, identified the operation and removed the account and organisation behind it.

It also said it had developed detection systems based on the campaign's behavioural signature.

The disclosure comes as Anthropic itself has increased its focus on election-related safeguards.

The company says its policies prohibit the use of Claude to run deceptive political campaigns, create fake digital content intended to influence political discourse, interfere with elections or spread misleading information about voting processes.