This study contributes to research on digital inequalities in the context of artificial intelligence by examining user perceptions of deepfake technology. We focus on the stratification of deepfake knowledge and attitudes towards deepfakes as critical elements of technology access. Based on a survey of 1,421 German internet users, we analyse the role of sociodemographic variables, digital skills, and personal innovativeness in predicting deepfake knowledge. We then examine the role of deepfake knowledge in users’ assessments of risks and potentials associated with the technology. Our results point to a generally low level of knowledge and a strong focus on risks in internet users’ perceptions of deepfakes. We find that age, gender and educational attainment predict knowledge about deepfakes. Digital skills, personal innovativeness, and social media use also positively relate to deepfake knowledge. This knowledge, in turn, is shown to play a role in users’ positive attitude towards the technology. While age plays only a minor role, female gender strongly relates to low knowledge and negative attitudes towards deepfakes. We thus find evidence of a sizeable gender divide in user access to the novel deepfake technology.