The manipulation of digital content is not a new phenomenon, but the application of machine learning to the creation of deepfakes has (i) radically improved the quality of output, (ii) slashed the resources required to produce realistic fakes at previously unimaginable scale, and (iii) “democratized” the process with user-friendly tools and services. The proliferation of increasingly realistic fabricated content presents numerous potential risks to individuals, organizations and societies. This report builds on the proceedings of an interdisciplinary expert workshop on deepfakes held by the EPFL International Risk Governance Center in September 2019. It suggests a simple framework for categorizing and prioritizing potential deepfake risks, before providing an overview of 15 potential technological, legal and societal responses.