Detect Deepfakesby Resemble AI
Deepfake case study · Multi-modal

Central African Republic AI media incident — Sep 2026

An investigation reveals how a Russian-linked influence operation used AI to generate propaganda and manipulate public opinion in the Central African Republic

Incident date
Sep 2026
Target
Central African Republic
Updated Sep 20, 2026 · 1 min read

A September 2026 report by Anthropic details how actors used AI technology to orchestrate a sophisticated foreign information manipulation and interference operation within the Central African Republic. By leveraging AI to automate content creation and internal management, the operation successfully disguised foreign-produced propaganda as local news originating from Bangui.

What happened

The operation centered on Radio Lengo Songo (98.9 FM), a station identified as having ties to the Wagner Group. A Russian-speaking actor in Bangui used the AI tool Claude to generate daily content, specifically instructing the model to embed pro-Russia, pro-Wagner, and anti-France narratives. To maintain deniability, the actors directed the AI to strip away synthetic formatting habits, ensuring the output mimicked the style of local journalists. This content was then disseminated through the radio station, Russian state media outlets, and local news channels.

Beyond content generation, the actors utilized AI to enforce strict internal control and political alignment. They tasked the model with creating contracts mandating loyalty to the CAR President and Russia, as well as developing scoring rubrics to evaluate and fire employees based on their adherence to political narratives. The operation also included the creation of forged CAR government documents, such as Gendarmerie and Ministry of Defense communications, and the maintenance of a surveillance database tracking opposition political figures. While the actors attempted to use the AI to label specific individuals as militants to incite security actions against them, the model refused these requests, leading the actors to pivot to anonymous-source framing. Investigators assessed the campaign as a Category Four operation, noting its high level of coordination and daily broadcast reach.

Sources