Identifying and mitigating the security risks of generative AI

The paper summarizes the results of a workshop held at Google and co-organized by Stanford University and the University of Wisconsin–Madison, studying the dual-use dilemma of generative AI. Generative AI technologies such as large language models and diffusion models can also be used by attackers to generate new attacks and to increase the speed and effectiveness of existing attacks. The paper discusses the community's short-term and long-term goals on this issue, hoping to provide the research community with a starting point for discussion and open questions.