Published: November 19, 2026 8:45 PM PST
Last Updated: November 19, 2026 8:45 PM PST
On 17 November 2026, researchers publicly disclosed a vulnerability in the OpenJDK Base64 decoder, tracked as CVE-2026-114938 and nicknamed "Java4Shell". The vulnerability affects Java 17, 21 and 25 running on x86-64 processors with AVX-512 Vector Byte Manipulation Instructions (VBMI): Intel Ice Lake or newer, and AMD Zen 4 or newer. The researchers report that it can be used for remote code execution in some widely deployed Java applications. Technical details and demonstrations are public, so we recommend acting now.
Affected products
-
Confluent Cloud (Confluent-managed services)
Confluent manages the Java runtime for Confluent Cloud. This issue has already been mitigated, and no further action is necessary. We have not seen any exploitation attempts of this vulnerability against our infrastructure or products.
-
Confluent Platform (Customer-managed components)
-
Products installed from packages or archives use the Java runtime you provide. They are affected when that Java runtime is version 17, 21 or 25:
- Confluent Platform 7.6.x to 8.4.x
- Control Center 2.0.x to 2.7.x
-
Docker images ship their own Java runtime, so they are affected as published:
- Confluent Platform 7.7.x to 8.4.x images
- Control Center 2.0.x to 2.7.x images
- Confluent Manager for Apache Flink (CMF) 2.0.x to 2.5.x
- Confluent Platform for Apache Flink images 1.19, 1.20, and 2.0 to 2.3 (-java17 and -java21 tags)
- Confluent Platform for Apache Flink Kubernetes Operator 1.12.x to 1.16.x
- Confluent Gateway 1.1.x to 1.4.x
- Confluent for Kubernetes (CFK) init container in CFK 2.10.x to 3.2.x
-
Products installed from packages or archives use the Java runtime you provide. They are affected when that Java runtime is version 17, 21 or 25:
Recommended actions
Confluent Platform (Customer-managed components) - Mitigations
We are preparing releases that ship a fixed Java runtime. Until they are available, apply this mitigation:
- Add -XX:-UseBASE64Intrinsics to every affected Java process running Java 17, 21 or 25, and restart it. Add it only to those processes: on Java 8 the process will not start with this flag.
- Confirm the flag on each running process with jinfo -flag UseBASE64Intrinsics <pid>; The output must be -XX:-UseBASE64Intrinsics. Checking the command line is not enough, because a later option can undo the flag.
Where to set the flag:
| Component | Configuration |
|---|---|
| Kafka broker, KRaft controller, ZooKeeper, Connect, Replicator | KAFKA_OPTS |
| Schema Registry | SCHEMA_REGISTRY_OPTS |
| REST Proxy | KAFKAREST_OPTS |
| ksqlDB | KSQL_OPTS |
| Control Center (legacy and next-gen) | CONTROL_CENTER_OPTS |
| Confluent Gateway | GATEWAY_OPTS (CFK: spec.podTemplate.envVars) |
| Confluent for Kubernetes |
spec.configOverrides.jvm: ["-XX:-UseBASE64Intrinsics"] on every Confluent Platform custom resource:
|
| cp-ansible |
Each component's *_custom_java_args:
Rerun the playbook with *_skip_restarts: false. |
| Package installs (systemd) | The same per-component variable through Environment=, then restart |
| Confluent Manager for Apache Flink, Confluent Platform for Apache Flink Kubernetes Operator and Flink jobs | See "Confluent Platform for Apache Flink" below. Do not use JAVA_TOOL_OPTIONS. |
If a variable is already set, append the flag at the end, separated by a space. Make sure nothing applied later re-enables it, such as -XX:+UseBASE64Intrinsics or _JAVA_OPTIONS.
Confluent Platform for Apache Flink
Add the flag in three places. Restarting Confluent Manager for Apache Flink or the operator does not restart running jobs, so redeploy every job afterwards.
-
Confluent Manager for Apache Flink: add the flag to the chart value jvmArgs, keeping existing arguments, and run helm upgrade with your current chart version. On version 2.5, if you customized cmf.statementService, also add the flag to:
- Any sharedCompilers entry that sets its own jvmArgs, and
- Both workers.jvmArgs and workers.compilerJvmArgs. Setting a worker value replaces its default, so keep a heap size in it.
- Kubernetes Operator: add the flag to jvmArgs.operator and jvmArgs.webhook. Under defaultConfiguration.flink-conf.yaml, set kubernetes.operator.default-configuration.flink-version.v1_19+.env.java.default-opts.jobmanager and .taskmanager to the flag. On chart 1.16 or later, use config.yaml instead if your values already set it. Do not use default-opts.all. Run helm upgrade, then restart the operator deployment.
-
Flink jobs: redeploy every job.
- Applications and FlinkDeployments: change spec.restartNonce.
- SQL statements on a dedicated compute pool: stop and resume each statement.
- SQL statements on a shared compute pool: stop the statements, change spec.clusterSpec.restartNonce, and resume them once the pool is running.
-
FlinkSessionJobs: redeploy the session cluster.
If a job sets its own env.java.default-opts.jobmanager or .taskmanager, add the flag to the end of that value. If a job sets -XX:+UseBASE64Intrinsics in env.java.opts.*, remove it.
To confirm, check that the last UseBASE64Intrinsics option on each running Confluent Manager for Apache Flink, operator, webhook, compiler, JobManager and TaskManager process is -XX:-UseBASE64Intrinsics. For version 2.5 compile workers, check that SP_JVM_OPTS and SP_COMPILER_JVM_OPTS in the pod spec contain the flag.
Performance
The -XX:-UseBASE64Intrinsics flag turns off the hardware-accelerated Base64 routine for both encoding and decoding, and Java uses its standard implementation instead. Base64 results are identical. The flag disables the hardware-accelerated code for both encoding and decoding, which may affect performance depending on your workload. We recommend testing to understand any performance impact in your environment.
Support
For help, open a case on the Confluent Support Portal and reference CVE-2026-114938 or Java4Shell.