Sandboxing Untrusted Workloads
A sandbox constrains files, network, CPU, memory, syscalls and credentials so untrusted computation cannot spend or reach everything.
Frame the problem
Security starts with a concrete asset, attacker capability and trust crossing.
Why the system fails
Isolation limits only one dimension or injects credentials that recreate broad authority from inside the sandbox.
The important question is not “what is Sandboxing Untrusted Workloads?” but “which assumption let untrusted data or an over-scoped identity cross untrusted workload → host and network capabilities?” Trace the decision at the boundary, then constrain what can happen after the first control fails.
Design the control in layers
Start with the control closest to the interpretation or privilege boundary: Deny filesystem and network by default Then add a control that reduces blast radius and telemetry that proves the decision was enforced.
The resulting design is not labelled secure. Record the identified controls, the known failure paths, the remaining exposure, and the evidence you would need during an incident.
| Prevent | Detect | Recover |
|---|---|---|
| Deny filesystem and network by default · Apply syscall, CPU, memory and process limits · Inject only task-scoped short-lived credentials | Sandbox policy violations and abnormal resource use | Contain the affected identity or component, scope impact from audit evidence, and preserve a regression test. |
Key points
- Asset: Host integrity, neighboring workloads, credentials and infrastructure cost.
- Boundary: Untrusted workload → host and network capabilities
- Primary control: Deny filesystem and network by default
- Detection signal: Sandbox policy violations and abnormal resource use
- Always ask what limits damage when the primary control fails.
Boundary control exercise
This lesson uses the shared boundary-control exercise.
Follow the attack
Safe conceptual simulation: capability → missing control → crossed boundary → asset impact.
- 1Attacker starts with: Untrusted uploaded code, parser input, build step or agent-generated command.
- 2Isolation limits only one dimension or injects credentials that recreate broad authority from inside the sandbox.
- 3The weak or missing boundary control is crossed: Untrusted workload → host and network capabilities
- 4Impact: Host access, data exfiltration or resource exhaustion.
- Host access, data exfiltration or resource exhaustion.
Defend, detect, recover
One prevention is a single point of security failure. Layer it and make failure observable.
- • Deny filesystem and network by default
- • Apply syscall, CPU, memory and process limits
- • Inject only task-scoped short-lived credentials
- • Sandbox policy violations and abnormal resource use
- • Contain the affected identity or component.
- • Scope access from audit evidence.
- • Fix the boundary and add a regression test.
- • Misconfiguration and new access paths can bypass the intended control.
- • A privileged insider or compromised control plane may still reach the asset.