This paper presents a systematic analysis of Twitter accounts using human faces generated by Generative Adversarial Networks (GANs) for profile pictures. The authors collected a dataset of 1,420 such accounts, which are used for various inauthentic activities such as spreading scams, spam, and amplifying coordinated messages. They developed a method to identify GAN-generated profiles by leveraging the consistent eye placement feature of GANs and human annotation. Applying this method to a random sample of active Twitter users, they estimate the prevalence of profiles using GAN-generated faces to be between 0.021% and 0.044%, or around 10,000 daily active accounts. The study highlights the emerging threats posed by multimodal generative AI and provides practical heuristics for social media users to recognize and defend against these accounts. The authors also release the dataset and code to facilitate further research.This paper presents a systematic analysis of Twitter accounts using human faces generated by Generative Adversarial Networks (GANs) for profile pictures. The authors collected a dataset of 1,420 such accounts, which are used for various inauthentic activities such as spreading scams, spam, and amplifying coordinated messages. They developed a method to identify GAN-generated profiles by leveraging the consistent eye placement feature of GANs and human annotation. Applying this method to a random sample of active Twitter users, they estimate the prevalence of profiles using GAN-generated faces to be between 0.021% and 0.044%, or around 10,000 daily active accounts. The study highlights the emerging threats posed by multimodal generative AI and provides practical heuristics for social media users to recognize and defend against these accounts. The authors also release the dataset and code to facilitate further research.