Paragraph 1
Artificial intelligence company Anthropic has disclosed that its Claude chatbot was weaponised to mass-produce fake social media posts praising Kenya’s Energy Cabinet Secretary Opiyo Wandayi while simultaneously attacking the country’s opposition, in what the firm described as a “local Kenyan political astroturfing campaign”. The AI safety and research company revealed that a single user generated Kenyan political content designed to appear as organic public sentiment, including batches of fabricated posts on the social media platform X. The content lavishly praised Mr Wandayi for allegedly halting a scheduled electricity tariff increase, while also pushing claims that the United Opposition coalition was fracturing ahead of the 2027 elections, specifically targeting prominent political figures including former deputy president Rigathi Gachagua and former president Uhuru Kenyatta. The disclosure, contained in a new Anthropic report detailing AI misuse cases on its platform between December 2025 and August 2026, underscores how AI is rapidly emerging as a potent new instrument for political influence in Kenya, as political operators increasingly deploy the technology to manufacture the appearance of widespread public support for favoured politicians.
Paragraph 2
Anthropic said it identified and removed an account used to orchestrate the operation, noting that the actor employed Claude across multiple sessions to systematically generate batches of exactly 50 tweets at a time. The prompts given to the chatbot explicitly instructed that the posts should resemble spontaneous, grassroots public commentary rather than a coordinated campaign, betraying a sophisticated understanding of how to evade detection. The operation relied on strategic hashtags including #PowerReliefKE and #PoweringTheNewKenya to amplify its messaging and create the illusion of a groundswell of popular support. The report stressed that the entire enterprise was built to look like organic civic engagement, with the generated content carefully crafted to mimic the tone, style, and concerns of ordinary Kenyan citizens expressing genuine gratitude toward government policy. The deliberate avoidance of any overt signs of coordination, combined with the high-volume output, demonstrates how generative AI tools are lowering the barrier for sophisticated disinformation operations that previously required substantial resources and manpower.
Paragraph 3
While Anthropic’s investigation could not definitively establish the identities of those behind the operation, the company noted that the pro-government tone and specific hashtags suggested it was “plausibly aligned” with the supporters of President William Ruto and the ruling coalition. The report referenced this as an organised effort that masquerades as a natural, spontaneous public movement but is in fact sponsored by hidden political interests. Notably, Anthropic said the operation did not ultimately reach real people, with its activity remaining isolated within a closed network of fake accounts and local influences on a single platform. The investigation was launched following a tip from rival US firm OpenAI about undisclosed criminal activity on its own ChatGPT platform, highlighting the growing intelligence-sharing among major AI companies. Anthropic emphasised a unique advantage of its position: whereas a social media site typically only sees a manipulation operation once content is already in circulation, AI providers may detect such activity on their platforms while the operation is still being constructed, offering a critical window for intervention.
Paragraph 4
This revelation arrives amid intensifying concern about the use of AI to manipulate political and civic discourse, particularly in nations where the information ecosystem is rapidly digitising. Kenya’s media landscape has become increasingly dependent on platforms such as Facebook, X, Instagram, TikTok, YouTube, and WhatsApp, creating expansive new avenues for political messaging that operate outside the traditional gatekeeping of television, newspapers, and radio. A recent study by the Kenya ICT Action Network (KICTANet), based on monitoring conducted between January and March 2026, found that alternative media actors on social media are actively shaping political narratives, often with considerably less editorial accountability than conventional media outlets. The study revealed a troubling pattern in which misleading narratives move across platforms, becoming progressively harder to challenge as they spread. The think tank specifically flagged Facebook as a starting point for fabricated newspaper headlines, doctored political graphics, and impersonated editorial content, before narratives migrate to TikTok where they are amplified through emotionally charged and selectively edited clips, and then sustained on YouTube and podcast channels through extended commentary and speculation.
Paragraph 5
Platform algorithms play a critical role in this amplification process, according to KICTANet, which observed that content generating stronger emotional reactions—whether anger, humour, outrage, fear, or political loyalty—tends to receive wider amplification through recommendation systems and reposting behaviour. This dynamic means sensational political narratives frequently receive greater visibility than balanced or evidence-based reporting, distorting public understanding of political issues. The growing integration of generative AI into this ecosystem dramatically heightens the risks, as experts warn that AI-generated deepfakes and manipulated content will become increasingly sophisticated as the 2027 elections approach. KICTANet cautioned that false content may become more convincing and harder to detect, that manipulated media may spread across platforms at unprecedented speed, and that verification processes may struggle to keep pace with the volume and quality of fabricated material. The combination of hyper-partisan algorithmic amplification and cheap, accessible AI generation tools creates a perfect storm for information disorder in Kenya’s political landscape.
Paragraph 6
Kenya has confronted politically driven online manipulation before. During the 2017 General Election, British political consulting firm Cambridge Analytica was embroiled in controversy over its use of data analytics, psychological profiling, social media advertisements, and grassroots networks to influence voters. The company was accused of harvesting Facebook user data through a third-party application to target users with fake-news campaigns, a scandal that triggered global outrage and regulatory scrutiny. The current Anthropic disclosure suggests that the tactics pioneered by such firms have now been democratised through AI tools, enabling even resource-constrained political operators to execute large-scale manipulation campaigns. As Kenya hurtles toward the 2027 General Election, the convergence of ubiquitous social media dependency, algorithmically amplified outrage, and accessible generative AI presents a formidable threat to the integrity of political discourse. The discovery of this operation, while ultimately unsuccessful in reaching genuine audiences, serves as an ominous warning that the tools for manufacturing public consent are becoming ever more powerful, ever more accessible, and ever more difficult to trace.
