Date created: | Last Updated:

Category: Project

Description: This project presents the results of an artificial intelligence (AI) audit conducted to explore the propensity of several large language model (LLM)-powered chatbots to reiterate claims and sources promoted by the Kremlin’s propaganda apparatus. We examine whether ChatGPT, Gemini, Copilot, and Grok return Kremlin-linked disinformation or reference Kremlin-linked disinformation sources in response to political questions, and explore both consistent and stochastic variation in their outputs. Based on 416 outputs generated in response to disinformation-related queries in the United Kingdom and Switzerland, we also explain why and under what conditions chatbots can return or reference Kremlin-linked disinformation. The UK and Swiss datasets, as well as the document describing preliminary results, are available under the tab ‘Files’.