Security Risks in Tool-Enabled AI Agents: A Systematic Analysis of Privileged Execution Environments

arXiv:2605.09721v1 Announce Type: cross Abstract: Tool-enabled AI agents are increasingly deployed in cloud-hosted environments and offered as services, where they perform side-effecting operations through privileged tools within execution environments. While such agents enable powerful automation,...