Back to Rules
Active Protection
VM Sandbox Hardening
Prevents Node sandbox abuse
Contributed by
Henry Lock
Detection Logic (Rego)
rule.regoOpen Policy Agent
deny if startswith(operation, "vm.") and not trusted_moduleThreat Context
When AI agents operate autonomously, they often require broad permissions to execute tasks. However, if an agent is compromised via prompt injection or hallucinates a destructive command, it can cause significant damage to the host system.
The VM Sandbox Hardening rule acts as a deterministic safeguard. It continuously monitors the agent's runtime behavior and immediately intercepts any action that violates this rule, ensuring your infrastructure remains secure regardless of the agent's internal state.
Remediation & Logs
When this rule is triggered, OpenLeash will block the execution and generate an audit log. You can view these events in your OpenLeash dashboard or forward them to your SIEM.
[BLOCKED] Rule Violation: VM Sandbox Hardening
Timestamp: 2026-03-20T12:11:45.768Z
Action: Execution Intercepted
Reason: Prevents Node sandbox abuse