Anthropic has uncovered a Kenyan political campaign that used its Claude AI system to produce large volumes of posts praising the government and attacking opposition figures.
The campaign was traced to a single domestic actor. The account was removed after the activity was detected. Anthropic said the operation was designed to make coordinated posts appear to come from ordinary Kenyans.
Posts Made to Look Organic
The campaign was identified as GTG-54004. Batches of 50 posts were generated at a time.
Specific political themes were supplied to Claude. Hashtags were also given. The material was then shaped to resemble spontaneous online commentary.
Anthropic said the aim was to create the appearance of public support. The campaign was therefore classified as coordinated inauthentic behaviour.
The company said the same user also ran commercial marketing work for Kenyan retail brands. Online identities including “SHANKI” and “Elkins Marketer” were linked to that activity.
Wandayi Became a Key Target
A major part of the campaign focused on Energy and Petroleum Cabinet Secretary Opiyo Wandayi.
Posts were produced in his favour. They followed his decision to halt a planned Kenya Power electricity tariff review.
The content was circulated with hashtags including #PowerReliefKE and #PoweringTheNewKenya, according to Anthropic.
The campaign also focused on the opposition. Claims about divisions within the United Opposition were promoted ahead of the 2027 General Election.
Former Deputy President Rigathi Gachagua was among those targeted. Former President Uhuru Kenyatta was also mentioned in the campaign.
No Government Link Was Found
Anthropic said the operation had a pro-government tone. But investigators did not establish direct involvement by the Kenyan government.
The company described it as a local campaign. Its exact organisers were not identified.
That distinction is important. The report does not establish that state officials ordered or financed the operation.
Instead, the findings point to a domestic actor seeking to shape political discussion through artificial online activity.
The Campaign Had Limited Reach
Despite the volume of material produced, the campaign appears to have had little effect on real voters.
Anthropic placed the operation in Category One on its Breakout Scale. The content remained largely within a network of fake accounts and local influencers on one platform.
The company said it found no evidence that the campaign reached a broad genuine audience.
That finding offers an important counterpoint. AI can produce political material quickly. But production alone does not guarantee influence.
A Wider AI Security Problem
The Kenyan case was one part of a much broader Anthropic investigation.
The September report examined misuse involving cyberattacks, espionage, surveillance, fraud, biological risks and weapons research. The cases covered activity identified between December 2025 and August 2026.
Anthropic said its systems had been used in increasingly complex operations. Some actors attempted to automate several stages of attacks.
The company said accounts linked to the abuse were removed. New detection methods were also developed from the cases identified.
A Warning for Kenya’s 2027 Race
The Kenyan case arrives as political activity builds ahead of the 2027 election.
Online platforms are already central to political campaigns. The use of AI adds another layer to that contest.
Synthetic posts can be produced in large numbers. They can also be tailored to specific audiences and political themes.
Yet the Anthropic findings also show the limits of such campaigns. A large volume of artificial content does not necessarily translate into public support.
For Kenya, the episode offers an early warning. Political debate is likely to become harder to separate from manufactured online activity as election campaigning intensifies.
About the Author
