How much you isolate a workload depends on how much you trust it. Syself Autopilot gives you several options, from the shared-kernel default up to a per-pod virtual machine, plus the per-pod controls that harden any container. This page maps those options and links to the page that configures each.

## Isolation levels

| Level                                       | Kernel           | Container root on the host | Runs on         |
| ------------------------------------------- | ---------------- | -------------------------- | --------------- |
| Standard runtime (default)                  | shared with host | host root (UID 0)          | every node      |
| + user namespace (`hostUsers: false`)       | shared with host | unprivileged host user     | every node      |
| Secure runtime (`runtimeClassName: secure`) | own guest kernel | isolated in a VM           | bare metal only |

- **Secure (Kata) runtime** puts each pod in its own lightweight VM with its own kernel, so a container or kernel escape stays trapped in the VM. Use it for untrusted or risky code. See [Container runtimes](/docs/hetzner/apalla/workloads/runtime-classes/container-runtimes).
- **User namespaces** remap container root onto an unprivileged host user, so an escape from a root container lands as an unprivileged user on the node instead of host root. Use it when an image must run as root and you cannot change it. See [Run a workload in a user namespace](/docs/hetzner/apalla/workloads/secure/run-in-a-user-namespace).

## Harden every workload

Whichever isolation level you pick, a production workload should also set its own security context: run as non-root, drop Linux capabilities, block privilege escalation, and keep the root filesystem read-only. Seccomp and AppArmor are already enforced by default. See [Set security contexts](/docs/hetzner/apalla/workloads/secure/set-security-contexts).

To enforce these rules across a whole namespace rather than per pod, apply [Pod Security Standards](/docs/hetzner/apalla/security/enforce-pod-security-standards).