9.4 Sandboxing

Description

Agent execution MUST be sandboxed to isolate it from the host system.

Rationale

Sandboxing limits blast radius. Even with code execution, attackers cannot access the host.

Audit Procedure

1. Check if agent runs in container/VM
2. Verify container security settings
3. Check for seccomp/AppArmor profiles

Remediation

1. Run in container with security settings
2. Use gVisor/Firecracker for code execution
3. Implement namespace isolation

Framework Mappings