Skip to content
Book Open access

AI Agentic Security: Safeguarding the Next Generation of Autonomous Systems

Nov 2026 · Proceedings of the 2026 ACM SIGSAC Conference on Computer and Communications Security · 0 citations

Abstract

The emergence of agentic artificial intelligence (AI)—systems capable of autonomous decision-making, goal-directed action, and interaction with complex digital and physical environments—represents a fundamental shift in AI. Securing these systems requires moving beyond the protection of individual models to the protection of continuously operating, tool-using entities that can act on their environments, collaborate with other agents, and adapt over time. These capabilities substantially expand the attack surface and introduce risks such as prompt and goal injection, memory poisoning, tool misuse, privilege compromise, manipulation of inter-agent communication, rogue-agent behavior, and software supply-chain attacks. This talk examines these emerging threats with reference to the OWASP Agentic AI threat model and considers how established security principles, including least privilege, defense in depth, complete mediation, and psychological acceptability, must be reinterpreted for agentic systems. Using the AgentSandbox architecture [1] as a case study, the talk illustrates how security-by-design can be incorporated through ephemeral task-specific agents, context minimization, granular privilege management, and policy-driven control of internal and external interactions. It concludes by identifying key research challenges, including formal guarantees for agent behavior, scalable assurance mechanisms, secure multi-agent ecosystems, and trust frameworks for agent identity, accountability, and governance. Addressing these challenges is essential if agentic AI is to evolve into a trustworthy foundation for autonomous systems at scale.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.