Anthropic shuts down Kenyan AI-driven propaganda campaign praising Gov't

Anthropic published one of its most detailed threat intelligence reports, detailing sophisticated attempts by external actors to misuse Claude, and outlining the defence measures taken to neutralise those threats.

According to a public update shared by the company, the comprehensive report exposes high-level, complex misuse attempts spanning critical domains, including cyberattacks, coordinated influence operations, domestic and international surveillance, biology, and weapons development.

Amongst thousands of such flagged projects was a Kenyan propaganda firm whose activities, Anthropic said, were malicious, inauthentic, misleading and flagrantly manipulative.

The AI research lab reported disrupting a coordinated inauthentic behaviour campaign originating in Kenya, where one user, described as "a single domestic actor", used Claude to mass-produce pro-government propaganda and attack opposition leaders.

The operation, tracked under the designation GTG-54004, involved generating batches of exactly 50 tweets designed to mimic organic, grassroots public sentiment.

"We identified and removed an account that was used by a single actor to mass-produce Kenyan political content. The operation was built to look like spontaneous, grassroots public sentiment," read the statement from Anthropic.

"The actor used Claude across several sessions to generate batches of exactly 50 tweets, explicitly instructing the model to make posts look like spontaneous grassroots commentary rather than a coordinated campaign."

According to Anthropic’s Threat Intelligence team, the Kenyan operator supplied specific political topics, talking points, and hashtags, instructing the AI model to humanise and format the posts into deployment-ready, cross-platform interactive widgets.

The primary focus of the propaganda network was shaping public perception around key national political issues.

A significant portion of the synthetic posts praised Energy Cabinet Secretary Opiyo Wandayi following his decision to halt a scheduled electricity tariff review by Kenya Power.

"The network focused its messaging on key Kenyan political issues and figures. A large portion of the AI-generated tweets praised Energy Cabinet Secretary Opiyo Wandayi for stopping a scheduled Kenya Power electricity tariff hike, using the hashtags #PowerReliefKE and #PoweringTheNewKenya to boost visibility," reads the expose.

Concurrently, the campaign sought to weaken political rivals by pushing narratives that Kenya's United Opposition coalition was fracturing ahead of the 2027 General Election.

The automated messaging directly targeted prominent opposition-aligned figures, including former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta.

"Additionally, the network pushed stories claiming that Kenya’s United Opposition coalition was breaking apart before the 2027 General Election, directly targeting politicians Rigathi Gachagua and former President Uhuru Kenyatta," Anthropic announced.

Anthropic’s investigation revealed that the operator utilised an identical AI workflow to run commercial marketing campaigns for Kenyan retail brands under the online persona "SHANKI" or "Elkins Marketer". 

While the network exhibited a pro-administration stance aligned with the ruling coalition, investigators confirmed there was no evidence of direct government involvement, classifying the activity as an entirely domestic, localised astroturfing effort.

"While we do not know the exact real-world identity of the people behind this operation, we believe this was a local Kenyan political astroturfing campaign. The network’s pro-administration tone and specific hashtags suggest it was plausibly aligned with the ruling coalition, though we have not identified the exact organisation responsible," the report noted.

"Based on a tip shared by OpenAI about recidivist activity on their platform, we conducted an internal investigation into suspected coordinated inauthentic behaviour in Kenya and identified this operation. We removed the account and the organisation behind it, and built detections around its behavioural signature," it stated.

Despite the structured nature of the operation, Anthropic evaluated its real-world impact using the Breakout Scale and classified it as Category One. 

The synthetic content remained completely isolated within a network of fake accounts and local influencers on a single platform, ultimately failing to reach or influence genuine voters.

The disclosure comes at a critical juncture for digital governance in East Africa, as the Kenyan government actively explores strategic partnerships with AI firms - including Anthropic - to build national AI strategies, enhance cybersecurity frameworks, and regulate emerging technologies. 

The incident highlights the growing challenge platforms face as domestic political operatives increasingly leverage generative AI to manufacture artificial public consensus at scale.

While Anthropic noted that these high-sophistication cases are not typical of general platform traffic, they highlight emerging vectors in AI misuse and point to where safeguards require ongoing refinement.

Anthropic emphasised that releasing the report serves a broader industry mandate. By disclosing detailed mechanics of these sophisticated threat operations, the company aims to help other platforms spot similar malicious activity on their own systems while offering the public a clearer, data-backed view of emerging security risks associated with advanced AI platforms.