Kenya: AI Influence Operation Targets 2027 Election

• An AI-driven influence operation, GTG-54004, targeted Kenya's 2027 General Election by promoting Opiyo Wandayi and discrediting Uhuru Kenyatta and Rigathi Gachagua.
• The operation used Anthropic’s Claude AI model to generate large volumes of social media posts, making them appear as spontaneous grassroots commentary.
• Narratives focused on portraying the United Opposition as fractured and promoting Wandayi's role in halting an electricity tariff increase.
• The same AI workflow was also used for commercial marketing campaigns, raising questions about repurposing techniques for political influence.
• Anthropic assessed the operation had minimal impact, failing to reach or influence real people beyond the network of fake accounts involved.

While this specific Kenya 2027 election AI influence operation proved ineffective, its discovery underscores the growing trend of generative AI in political communication.

A sophisticated artificial intelligence-driven influence operation, identified as GTG-54004, was uncovered targeting the political landscape ahead of Kenya's 2027 General Election. This covert social media campaign aimed to manipulate public perception by discrediting key opposition figures while simultaneously promoting a government official. The findings, detailed in a threat intelligence report by global AI firm Anthropic, revealed that the operation specifically targeted former President Uhuru Kenyatta and former Deputy President Rigathi Gachagua with negative narratives, while promoting Energy Cabinet Secretary Opiyo Wandayi.

The GTG-54004 operation leveraged artificial intelligence to mass-produce social media posts, with a single operator utilizing Claude AI across multiple sessions. The operator explicitly instructed the AI model to generate batches of exactly 50 tweets, ensuring the content mimicked spontaneous grassroots commentary rather than a centrally organized campaign. These AI-generated narratives sought to portray Kenya's United Opposition as fractured and on the brink of collapse in the lead-up to the 2027 elections, directly targeting politicians Rigathi Gachagua and former President Uhuru Kenyatta.

Conversely, the campaign pushed favorable messaging for Opiyo Wandayi, particularly highlighting his role in preventing a scheduled electricity tariff increase. Hashtags such as #PowerReliefKE and #PoweringTheNewKenya were used to amplify these positive messages, with the AI-generated posts presented as independent commentary from ordinary social media users. The operator's workflow involved supplying Claude with pre-drafted topics, talking points, and hashtags, then instructing the model to refine the content for a more natural and human-like appearance. Intriguingly, the report noted that the identical AI-assisted workflow was also employed for commercial marketing campaigns in Kenya under the persona 'SHANKI'/'Elkins Marketer,' suggesting a repurposing of techniques developed for advertising into political influence operations.

Despite the sophisticated use of AI and the volume of generated material, Anthropic assessed that the GTG-54004 operation had minimal impact beyond the network of accounts involved. The activity was classified as Category One on Anthropic’s Breakout Scale, indicating it was entirely isolated within fake accounts and local influencers on a single platform, failing to reach or influence any genuine individuals. This assessment places the operation at the lowest level of impact, signifying its inability to penetrate wider public discourse or meaningfully sway real users.

While this specific Kenya 2027 election AI influence operation proved ineffective, its discovery underscores the growing trend of generative AI in political communication. The incident highlights the technology's capacity to produce large volumes of personalized messages at a speed and scale previously unattainable by conventional human-run campaigns. In response to the findings, Anthropic has banned the accounts implicated in the operation and implemented additional behavioral detections to identify similar activities, acknowledging the evolving challenge of AI in political campaigns in Africa and globally.

This report highlights the increasing use of generative AI in political influence operations, even if ineffective in this specific instance. Lawyers and compliance officers should advise clients on the potential for AI-generated disinformation, the importance of robust digital due diligence in campaigns, and the evolving regulatory and ethical considerations surrounding AI misuse in public discourse.

Source: Original reporting via Capital News