Anthropic Disrupts Kenyan AI-Powered Political Propaganda Campaign

NAIROBI, Kenya — Artificial intelligence safety company Anthropic says it has disrupted a Kenyan political influence operation that used its Claude AI model to generate large volumes of pro-government content and attack opposition figures ahead of the 2027 General Election.

The operation, identified by Anthropic as GTG-54004, involved what the company described as a single domestic actor using Claude across multiple sessions to produce batches of political posts designed to appear as spontaneous grassroots commentary.

Anthropic disclosed the operation in one of its most detailed threat intelligence reports, which examines attempts to misuse advanced AI across a range of areas, including coordinated influence operations, cyberattacks, surveillance, biology and weapons development.

Claude Used to Generate Political Content at Scale

According to Anthropic's Threat Intelligence team, the Kenyan operator repeatedly asked Claude to generate batches of exactly 50 posts, with instructions to make the material appear organic and independently produced.

The operator reportedly provided political topics, talking points and hashtags before asking the AI model to humanise and format the content for publication.

Anthropic said the campaign was designed to create the impression of genuine public support rather than reveal that the posts were part of a centrally coordinated operation.

The company described the activity as coordinated inauthentic behaviour, a term used to describe organised efforts to manipulate public discourse by disguising the source or nature of political messaging.

Campaign Praised Wandayi and Targeted Opposition

A significant portion of the campaign's content focused on Energy Cabinet Secretary Opiyo Wandayi, according to Anthropic.

The company said the operation generated posts praising Wandayi following his intervention over a planned Kenya Power electricity tariff review. The messaging reportedly promoted hashtags including #PowerReliefKE and #PoweringTheNewKenya.

At the same time, Anthropic said the campaign attempted to undermine opposition political figures by promoting claims that Kenya's United Opposition coalition was breaking apart ahead of the 2027 elections.

Among those targeted were former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta.

Anthropic said the political messaging was intended to shape public perceptions of key national issues and political figures.

No Evidence of Direct Government Involvement

Although the campaign adopted a pro-administration tone, Anthropic said it found no evidence linking the operation directly to the Kenyan government.

The company said the available evidence pointed to a local Kenyan astroturfing campaign whose messaging appeared plausibly aligned with the ruling coalition. However, investigators said they had not established the identity of the organisation behind the operation.

The distinction is significant because the findings do not establish that government officials or institutions commissioned, funded or directed the campaign.

Anthropic said the operation appeared to have been conducted by a domestic actor rather than being part of a foreign influence campaign.

Commercial Marketing Workflow Repurposed for Politics

Anthropic's investigation also uncovered links between the political activity and commercial marketing work.

According to the company, the operator used a similar AI-assisted workflow for marketing campaigns involving Kenyan retail brands while operating online under the names “SHANKI” and “Elkins Marketer.”

The finding illustrates how tools and techniques developed for legitimate digital marketing can potentially be adapted to political influence campaigns.

Generative AI can allow operators to rapidly produce large quantities of tailored content, alter the tone of messages and create posts designed for different audiences.

Anthropic Says Campaign Failed to Reach Genuine Voters

Despite the scale and structured nature of the operation, Anthropic assessed its real-world impact as limited.

The company classified the activity as Category One on its Breakout Scale, saying the synthetic content remained largely confined to a network of fake accounts and local influencers on a single platform.

As a result, Anthropic said the operation did not successfully reach or influence genuine voters.

The assessment suggests that while generative AI can significantly increase the volume of political content an operator can produce, the technology does not automatically guarantee that such content will reach large or authentic audiences.

OpenAI Tip Helped Trigger Investigation

Anthropic said the investigation began after it received a tip from OpenAI regarding suspected repeat activity on its platform.

Following the tip, Anthropic conducted an internal investigation into coordinated inauthentic behaviour in Kenya.

The company said it subsequently removed the account and organisation associated with the operation and developed new detection mechanisms based on the campaign's behavioural patterns.

The incident highlights the increasing cooperation among major AI companies as they attempt to identify and disrupt malicious use of generative AI across platforms.

AI and Kenya's 2027 Election

The disclosure comes as Kenya approaches the 2027 General Election, placing renewed attention on the role of artificial intelligence and social media in political communication.

The emergence of AI-generated political content creates a new challenge for voters, journalists, technology platforms and election authorities. Large volumes of synthetic posts can make it increasingly difficult to distinguish genuine public sentiment from organised attempts to manufacture the appearance of popular support.

The Kenyan case also demonstrates how domestic actors can potentially use widely available AI tools to scale political messaging without requiring a large team of human content creators.

For platforms, the challenge is therefore not limited to detecting false information. It also involves identifying behavioural patterns that indicate coordinated attempts to manipulate the appearance of public opinion.

Anthropic Strengthens Claude's Safety Measures

Anthropic said lessons from the Kenyan investigation and other threat incidents have been incorporated into Claude's safety infrastructure.

The company said the sophisticated cases outlined in its report are not representative of normal platform activity, but they demonstrate emerging methods of AI misuse that require continued monitoring and improved safeguards.

By publishing details of the operations, Anthropic said it hopes to help other technology companies identify similar patterns and strengthen their own defences.

The Kenyan case offers an early warning of how generative AI could influence future political campaigns: not necessarily by inventing an entire political strategy, but by allowing a relatively small number of actors to produce, refine and distribute persuasive messaging on a scale that was previously far more difficult to achieve.

With Kenya's 2027 election approaching, the ability to distinguish authentic public debate from AI-assisted political manipulation is likely to become an increasingly important issue for the country's digital information ecosystem.