Search for: All records

Misinformation on social media poses significant societal challenges, particularly with the rise of large language models (LLMs) that can amplify its realism and reach. This study examines how adversarial social influence generated by LLM-powered bots affects people’s online information processing. Via a pre-registered, randomized human-subject experiment, we examined the effects of two types of LLM-driven adversarial influence: bots posting comments contrary to the news veracity and bots replying adversarially to human comments. Results show that both forms of influence significantly reduce participants’ ability to detect misinformation and discern true news from false. Additionally, adversarial comments were more effective than replies in discouraging the sharing of real news. The impact of these influences was moderated by political alignment, with participants more susceptible when the news conflicted with their political leanings. Guided by these findings, we conclude by discussing the targeted interventions to combat misinformation spread by adversarial social influences.