Agent Sandboxing

Agent sandboxing is the practice of running an AI agent in an isolated environment with restricted access to systems, networks, and data. It reduces the blast radius when an agent is compromised, misconfigured, or behaving unpredictably.