A preprint of the paper titled “Anatomy of an AI-powered malicious social botnet” by Yang and Menczer was posted on arXiv. Concerns have been raised that large language models (LLMs) could be utilized to produce fake content with a deceptive intention, although evidence thus far remains anecdotal. This paper presents a case study about a coordinated inauthentic network of over a thousand fake Twitter accounts that employ ChatGPT to post machine-generated content and stolen images, and to engage with each other through replies and retweets. ChatGPT-generated content promotes suspicious crypto and news websites and spreads harmful comments. While the accounts in the AI botnet can be detected through their coordination patterns, current state-of-the-art LLM content classifiers fail to discriminate between them and human accounts in the wild. These findings highlight the threats posed by AI-enabled social bots and have been covered by Tech Policy Press, Business Insider, Wired, and Mashable, among others. And to no one’s surprise, versions of these articles likely summarized by ChatGPT already appear on plagiarized “news websites.”
CNetS alumnus 
Indiana University’s Observatory on Social Media, funded in part last year with a $3 million grant from the 
