Generative AI can support useful work while also making existing attacks easier or enabling new ones. This paper synthesizes findings from a workshop organized by Google, Stanford University and the University of Wisconsin–Madison.
The report examines this dual-use problem across technologies including large language models and diffusion models. It outlines risks, possible mitigations and research priorities over both short and longer time horizons. Rather than claiming an exhaustive account, it provides a starting point for security researchers investigating how these capabilities change the threat landscape.
