Self-State Attacks on Self-Hosted AI Agents: How Far Can OS Defenses Go?
This paper examines whether operating-system defenses can prevent self-state attacks that compromise self-hosted AI agents, highlighting a potential security gap in agent deployments.
This paper examines whether operating-system defenses can prevent self-state attacks that compromise self-hosted AI agents, highlighting a potential security gap in agent deployments.
Auto-scored: The paper identifies a real escalation vector for attacking AI agents, while also providing a partial defensive analysis.