Search K
Appearance
Appearance
This release includes updates in the following areas for the F5 AI Security Operator:
| Distribution | Version |
|---|---|
| Helm chart (f5-ai-security-operator-helm) | 1.8.0 |
| Red Hat Ecosystem Catalog (operator bundle) | 0.12.0 |
WARNING
GPU Operator v25.10.0+ enables Container Device Interface (CDI) mode by default. On OpenShift/RHCOS nodes running CRI-O with crun older than 1.26, this causes inference pods with a read-only root filesystem to fail to start. The root cause is a known crun issue, fixed by crun PR #1925.
Upgrading to OCP 4.21.28 or later resolves this (ships crun 1.27+). Check your node's crun version with:
oc debug node/<node> -- chroot /host crun --versionBecause crun is provided and managed as part of the OpenShift/RHCOS platform, work with Red Hat to upgrade to a supported OpenShift/RHCOS release that includes this fix. Do not manually replace the crun binary on production OpenShift nodes.
If you can't upgrade OCP yet, disable CDI in ClusterPolicy as a workaround:
oc patch clusterpolicy gpu-cluster-policy --type=merge -p '{"spec":{"cdi":{"enabled":false}}}'This release has the following changes to default behavior:
Hardened, restricted-v2-compatible inference images
The guardrails, red team, and remediate model images have been hardened (dropped capabilities, non-root execution). They now run under OpenShift's built-in restricted-v2 SCC. The operator no longer creates or requires a custom SCC for inference pods.
This release resolves the following CVEs in operator dependencies:
golang.org/x/net.golang.org/x/crypto.go.opentelemetry.io/otel.oras.land/oras-go/v2.The operator's Go toolchain has also been upgraded from 1.24 to 1.25.