Advisory review: 18 November 2026. Vendor guidance and patch availability may change; check the linked advisories before deploying.
What is Java4Shell?
Java4Shell is a memory corruption flaw in OpenJDK’s hardware-accelerated Base64 decoder. An application that sends attacker-controlled input into the affected implementation may expose its Java process to crashes or remote code execution. Red Hat identifies unpatched OpenJDK 17.0.6 and later builds, including the 21 and 25 lines, on x86-64 systems exposing AVX-512 VBMI. ARM64 and x86-64 systems without that feature are not known to be affected. Red Hat security bulletin.
Oracle assigns CVSS v3.1 8.1 (High); Red Hat classifies the impact as Critical. Oracle’s affected-product matrix also lists particular GraalVM builds and a specialized Java 8 performance build. Avoid a blanket claim that every Java 8 distribution is unaffected, or that a major Java version alone establishes patch status. Match the vendor and full build against its advisory. Oracle November 2026 risk matrix.
Why Keycloak needs immediate attention
Keycloak reports demonstrated pre-authentication remote code execution and affected runtimes in x86-64 images 26.8.2-1 and 26.8.2-2. CPU features and JVM settings still determine exposure. Its 18 November update names no fixed image. Keycloak advisory.
For prioritization, treat an identity service as a dependency of every application that trusts it. Assign one remediation owner and ask application owners to help identify deployments, including disaster recovery systems. Keep a record of which instances have been checked and which remain unknown.
Check the deployment that actually serves traffic
For each service, record the runtime vendor and full build, image digest, execution architecture, CPU features, and running JVM flags. Include bundled runtimes and containers. The Canadian Cyber Centre recommends prioritizing internet-facing systems and reviewing third-party Java products. Its 17 November alert reported no known active exploitation at that time; that statement is time-bound. Canadian Cyber Centre alert.
On a Linux execution node, these read-only checks are useful:
java -version
lscpu
grep -m1 '^flags' /proc/cpuinfo
Look for avx512vbmi. Use the Java executable belonging to the service; an interactive shell may select another installation. For a Keycloak container:
docker exec <keycloak-container> java -version
This starts a separate process, not a query of the active server’s flags. Check every eligible node, including failover capacity. Keycloak verification guidance.
Apply an interim mitigation when patching must wait
Red Hat recommends disabling Base64 intrinsics with:
-XX:-UseBASE64Intrinsics
Its alternative is -XX:UseAVX=2, which disables AVX-512 more broadly. Apply that alternative only on x86-64; Red Hat warns that ARM64 JVMs do not recognize it and can fail to start. Restart affected processes after changing startup configuration. These are interim measures pending a fixed runtime. Red Hat mitigation guidance.
For Keycloak’s server startup script, append the first flag to existing JAVA_OPTS_APPEND. Configure the service manager or deployment manifest that actually launches it, then restart all instances. Verify the live process:
jcmd <keycloak-pid> VM.flags -all | grep -F UseBASE64Intrinsics
Expect UseBASE64Intrinsics = false. The Operator is a separate JVM: its direct Java launcher uses JDK_JAVA_OPTIONS. Mitigating it does not configure managed Keycloak servers. Follow the advisory’s deployment-specific examples. Keycloak mitigation instructions.
Patch, redeploy, and collect closure evidence
Install a supported build whose vendor explicitly confirms the CVE fix. Microsoft recommends updating the runtime or the third-party product that bundles it; WAF mitigations provide supplementary protection while updates are applied. Microsoft Java4Shell guidance.
Use these acceptance criteria for your remediation work:
- Runtime evidence: retain the vendor advisory and the full deployed build identifier.
- Rollout evidence: record replacement image digests and confirm every replica was replaced, including standby instances.
- Process evidence: capture effective flags when relying on a temporary mitigation.
- Service evidence: exercise normal login, token refresh, logout, and federation flows used by your organization.
- Rollback evidence: ensure rollback artifacts have equivalent protection before marking the change complete.
Review unexpected JVM exits and unusual authentication activity alongside your normal incident response process. A crash alone does not prove exploitation, and a quiet log does not prove safety. If compromise is suspected, preserve evidence and assess exposed credentials with your incident responders.
Record an owner and deadline for any deployment still awaiting a permanent patch. For a practical way to document evidence and verification status, use our pentest report checklist.