Dagstuhl-Seminar — Autonomous AI Agents in Computer Security

AI is transforming computer security in ways that have significant potential to disrupt the effectiveness of current defensive standards. The vulnerabilities of the AI systems themselves, along with their potential use as offensive tools, introduce a new dimension to the threat modeling of computer systems. As AI models increasingly power autonomous agents, this attack surface continues to expand, giving rise to novel threats that impact both society and industry. At the same time, AI models and autonomous agents could also serve as transformative tools for developing advanced defensive techniques. The rapid pace of AI development creates significant uncertainty about what will be required to defend against AI-based attacks and how to devise autonomous cyber defenses. By reducing reliance on manual intervention and integrating with existing defense mechanisms, AI defensive systems may offer new ways to counter attacks and mitigate threats. This Dagstuhl Perspectives Workshop aims to develop a systematic understanding of how AI agents can be used in both offensive and defensive security.