A study linked to Meta’s Oversight Board is raising alarms about how easily AI chatbots could be used as propaganda tools. The report suggests modern bots may be uniquely effective at shaping narratives because they can generate polished, persuasive text at scale while tailoring responses to different audiences.
One of the most striking findings is that chatbot answers can change depending on the language used. According to the study, ChatGPT described China as not being a democracy when asked in English, but gave a more cautious answer in Chinese. That kind of variation points to how AI systems may present politically sensitive topics differently based on context.
The research also highlighted uneven responses to prompts about public figures and leaders. In the examples cited, Anthropic’s Claude would help create critical material about President Donald Trump or Britain’s King Charles III, while appearing to draw the line differently for other monarchies and ruling figures. The contrast suggests that AI guardrails may not be applied consistently across countries or political systems.
Taken together, the findings add to growing concerns that chatbots are not just neutral tools for information. If their outputs shift by language, region or subject matter, they could be used to amplify state narratives, limit criticism in sensitive areas or subtly influence public opinion in ways that are hard for users to detect.
