Critical Security Advisory: CVE-2026-114938 "Java4Shell" Java Vulnerability (Action Recommended)
We're notifying you of a critical vulnerability (CVE-2026-114938, "Java4Shell") in the Java runtime used by the AtScale engine, Keycloak, and atscale-monitor. It affects all AtScale Container releases and could allow remote code execution without authentication on some x86 servers. An interim mitigation that only needs a Helm values change is below.
Important details
Security Advisory: Java Base64 decoder vulnerability "Java4Shell" (CVE-2026-114938)
Severity: Critical (rated Critical by Red Hat; CVSS 3.1 score 8.1)
Affected: All AtScale Container releases
Resolved in: Future AtScale releases, as upstream Java fixes become available. We'll update this article when fixed releases ship.
Summary
A critical vulnerability (CVE-2026-114938) has been found in the Base64 decoder built into Java. A malformed Base64 string can overwrite memory in the Java process, which could let an unauthenticated attacker read memory or run code on the server.
Three AtScale components run on an affected Java version: the engine, Keycloak, and atscale-monitor. Keycloak carries the highest risk because it reads Base64 data from login requests before a user signs in.
There's no known exploitation of this vulnerability as of November 18, 2026.
Am I affected?
An attack only works when both of these are true:
- The component runs Java 17.0.6 or later. This is true for all AtScale Container releases.
- The Kubernetes node has an x86-64 CPU that supports AVX-512 VBMI (Intel Ice Lake and newer, AMD Zen 4 and newer).
ARM64 nodes, including AWS Graviton, aren't affected.
The mitigation below is safe on any CPU, so we recommend applying it to every AtScale cluster rather than checking each node.
How to check your CPU (optional)
You can check whether your nodes have an affected CPU without needing access to the nodes themselves. This command reads the CPU details through the AtScale keycloak pod. Replace <namespace> with your AtScale namespace and pod with the name of your keycloak pod:
#Execute the following command to get the name of your keycloak pod
kubectl get pods -n <namespace> |grep -i keycloak
#Execute the following command using the pod name from the previous output
kubectl -n <namespace> exec <keycloak_pod_name> -- cat /proc/cpuinfo | grep -o -w -m1 avx512vbmi
Read the result:
- If it prints
avx512vbmi, the node’s CPU is affected. Apply the mitigation below. - If it prints nothing, the node’s CPU isn’t affected.
This only checks the node the engine pod runs on. To see which node each AtScale pod runs on, look at the NODE column of this command:
kubectl -n <namespace> get pods -o wide
If your AtScale pods run on more than one node, or your cluster mixes instance types, the safest choice is to apply the mitigation to all components.
If you have shell access to a node, you can run the check on it directly:
grep -o -w -m1 avx512vbmi /proc/cpuinfo
Interim mitigation
The mitigation turns off the vulnerable Java decoder with one JVM flag: -XX:-UseBASE64Intrinsics. Java falls back to its standard decoder, which isn't affected.
Step 1: Add the flag to your Helm values file
Defaults differ between releases, so first look up your current JAVA_OPTS. Run this command and look for JAVA_OPTS under atscale-engine.options and atscale-monitor.options:
helm get values <release> -n <namespace> --all
Then add the flag to the end of each value in your values file:
keycloak:
extraEnvVars:
- name: JDK_JAVA_OPTIONS
value: "-XX:-UseBASE64Intrinsics"
atscale-engine:
options:
JAVA_OPTS: "<keep your existing options> -XX:-UseBASE64Intrinsics"
atscale-monitor:
options:
JAVA_OPTS: "<keep your existing options> -XX:-UseBASE64Intrinsics"
Please check these points before you apply it:
- atscale-engine & atscale-monitor
- Keep your existing
JAVA_OPTS. For the engine and atscale-monitor, add the flag to the end of the current value. Replacing the value removes the default memory and garbage-collection settings. If you've customizedJAVA_OPTS, add the flag to the end of your custom value. - keycloak
- Merge with existing entries. Helm replaces lists instead of merging them. If your values file already has
keycloak.extraEnvVarsentries, add the new entry to that list. - Use
JDK_JAVA_OPTIONS. Don't setJAVA_OPTS_APPEND. The AtScale chart already uses it for Keycloak cluster settings, and overriding it breaks caching across Keycloak replicas.
Step 2: Apply the change
Run your usual helm upgrade command with the updated values file.
helm upgrade <release> oci://docker.io/atscaleinc/atscale --version <your current version> -n <namespace> -f values.yaml
If you upgrade with --reuse-values and --set instead of a values file, set the whole keycloak.extraEnvVars list with --set-json, including any entries you already have. An indexed --set, such as keycloak.extraEnvVars[0].value, fails the upgrade. For example:
--set-json 'keycloak.extraEnvVars=[{"name":"JDK_JAVA_OPTIONS","value":"-XX:-UseBASE64Intrinsics"}]'
The engine, Keycloak, and atscale-monitor pods will restart. Keycloak's restart briefly interrupts user logins, so we recommend a short maintenance window.
Step 3: Confirm the flag is active
Each JVM logs a line at startup when it picks up the flag. This example assumes the release is named atscale. Adjust the resource names if yours differs.
for r in deploy/atscale-engine statefulset/atscale-keycloak statefulset/atscale-monitor; do echo "$r: $(kubectl -n <namespace> logs $r 2>/dev/null | grep -m1 'Picked up JDK_JAVA_OPTIONS')"; done
Each line should include -XX:-UseBASE64Intrinsics. For the engine, you can also ask the running JVM directly (the atscale-monitor image doesn’t include jcmd). It should print UseBASE64Intrinsics = false {product} {command line}.
kubectl -n <namespace> exec deploy/atscale-engine -c engine -- jcmd 1 VM.flags -all | grep UseBASE64Intrinsics
Performance impact
The flag -XX:-UseBASE64Intrinsics turns off one hardware optimization, the accelerated Base64 encoding and decoding in the Java runtime. Java falls back to its standard Base64 code, which is still compiled and optimized. No other JVM optimizations are affected.
We don’t expect a measurable effect on AtScale. The extra cost depends on payload size and can be measured in milliseconds, so this falls well within normal variation. Larger Base64 payloads decode more slowly than before, but the difference only becomes noticeable at tens of kilobytes, which isn’t a typical AtScale workload.
ARM64 nodes such as AWS Graviton aren’t vulnerable, so the flag isn’t required there. It’s safe to apply anyway if you’d rather use one configuration for all node types.
After you upgrade
Once you're running an AtScale release that includes the fixed Java runtime, the flag is no longer needed and will restore original performance. You can leave it in place or remove it.
References
- NVD: CVE-2026-114938
- OpenJDK Vulnerability Advisory, November 17, 2026
- CISA alert, November 17, 2026
Questions
If you need help applying the mitigation or planning an upgrade, please contact support.