AI chatbots amplify Russian propaganda, study finds

A research project by the British think‑tank Demos examined how large language models from leading tech companies may convey content from Russian propaganda sources. The study tested five AI systems – OpenAI's ChatGPT, Google Gemini, xAI, Anthropic and Mistral AI – using prompts taken from the so‑called r‑FBI foundation, a fake organization founded by the late Wagner commander Yevgeny Prigozhin and listed under EU and US sanctions. In 16.6 % of cases the models presented the propaganda narratives as credible, and in 30.9 % of responses the sanctioned source was not disclosed. About 52 % of answers contained caveats or denials of the claims. Carl Miller, co‑founder of Demos' Centre for Analysis of Social Media, warned that “these stories have no place in serious public discourse,” and cautioned that hybrid‑war actors could exploit generative‑search optimisation to shape AI‑generated information.

The findings highlight a risk that generative AI could be weaponised to spread disinformation, prompting calls for better safeguards and transparency in AI deployments.