Anthropic Report Details Claude Misuse for Kenyan Political Disinformation

Kenya was highlighted in Anthropic's recent global threat intelligence report due to a pro-government operator utilizing Claude models to create fake grassroots social media content in anticipation of the 2027 general election. [1] The details are outlined in the “Detecting and Countering Misuse of AI: September 2026” report, which covers malicious activities disrupted by the company from December 2025 through August 2026. [1] The report notes that an account was identified and removed after a single actor used it to mass-produce Kenyan political content intended to mimic spontaneous public sentiment. [1]

The actor employed Claude over multiple sessions to generate groups of 50 tweets, with explicit instructions for the model to ensure the posts appeared as organic commentary instead of a coordinated effort. [1] A significant portion of these AI-generated tweets praised Energy Cabinet Secretary Opiyo Wandayi for stopping a planned Kenya Power electricity tariff increase, utilizing hashtags like #PowerReliefKE and #PoweringTheNewKenya to increase visibility. [1] Additionally, the operation promoted narratives suggesting the United Opposition coalition was fracturing before the 2027 election, specifically targeting politicians such as Rigathi Gachagua and former president Uhuru Kenyatta. [1]

Beyond political messaging, Anthropic reported that the same actor used an identical process to market Kenyan retail brands. [1] This marketing effort was conducted under the persona “SHANKI”/”Elkins Marketer” and involved repackaging promotional broadcasts as organic tweets, with links included in every fifth post. [1] Anthropic stated there was no evidence of government involvement in these activities. [1] The company assessed the operation as being consistent with a domestic Kenyan initiative, although it did not name the specific organization responsible. [1] The campaign was described as having a pro-administration tone and using specific hashtags that aligned with those narratives. [1]

Anthropic's report provides concrete evidence of how advanced language models are currently being operationalized for domestic influence operations ahead of national elections. The specific use of Claude to generate batches of 50 tweets designed to look like organic commentary demonstrates the practical mechanics of astroturfing campaigns. While the primary focus was political, the actor's dual use of the same workflow for retail marketing highlights the commercial viability and crossover skills involved in these operations. What the sources don't address: whether the retail marketing activities were simply a front or if the operator functioned as a generic social media management service for multiple clients.