Oracle patches Java4Shell, the critical Base64 decoder flaw that executes code remotely without authentication
On November 17, 2026, Oracle shipped a fix for CVE-2026-114938, a memory-corruption bug in OpenJDK’s Base64 decoder that is exploitable without authentication against Keycloak, Jira, and Jenkins. If you run JDK 17.0.6 or newer on a recent x86-64 server, apply the patch or set -XX:-UseBASE64Intrinsics now.
November 17, 2026. Oracle ships a fix for CVE-2026-114938, a memory-corruption bug in the Base64 decoder of OpenJDK. November 17, 2026. The researchers at Calif, who found it, name it Java4Shell and rate it 9.2 under CVSS v4, against 8.1 under CVSS v3 from Oracle. November 17, 2026. CISA publishes an alert urging organizations to patch immediately. Why it matters: the flaw lives in the Java runtime itself, not in an application, and the vulnerable code runs before authentication — Calif’s demos show remote code execution against default Keycloak, Jira, Confluence, Artifactory, Jenkins, and PingFederate deployments.
A memory-corruption bug in the Base64 decoder’s assembly
Java4Shell needs no flawed library and no vulnerable framework. The bug sits in the handwritten assembly that speeds up Base64 decoding with AVX-512 instructions. A very short malformed input makes that code write 62 bytes past the end of a buffer, before Java even rejects the input.
The decisive point is what Base64 carries. The format transports authentication headers, tokens, and file uploads — which means the vulnerable code often runs before a user is authenticated. An attacker with no credentials can reach the decoding path, corrupt heap memory, read secrets out of it — private keys, tokens, personal data — and then chain into code execution. CISA adds that a crafted input can also crash the JVM, so availability is at risk alongside confidentiality.
The discovery itself is unusual. Calif says it found the bug with AI assistance in about an hour, and had a first working exploit two hours later, for under $200 in AI token costs. The researchers are holding their product-specific exploits while organizations catch up, but they say they will publish them, and the root cause is already public. The cost of replication has dropped: nobody should plan around the absence of a public exploit.
The scope: recent JDKs on recent CPUs
A system is in scope only when two conditions are both true. On the runtime side: any OpenJDK build from 17.0.6 onward — the 17, 21, 25, and 27 lines, from any vendor — Oracle, Amazon Corretto, Eclipse Temurin, Azul Zulu, Red Hat, Microsoft, SAP. One exception on Java 8: Oracle lists the 8u511b33-perf build of the Java SE 8 Enterprise Performance Pack as vulnerable. Releases before 17.0.6 are unaffected.
On the processor side: only x86-64 machines with the AVX-512 VBMI extension — Intel Xeon from Ice Lake through Granite Rapids, and AMD EPYC/Ryzen on Zen 4 and newer. ARM64 and x86-64 without AVX-512 VBMI are out of scope.
One command settles the hardware question on Linux: the host must expose all four flags.
grep -m1 '^flags' /proc/cpuinfo | tr ' ' '\n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$' Two scoping traps to keep in mind. First, check the runtime each application actually runs: many products and container images bundle their own JDK, which operating-system updates never touch. Second, use the CPU condition to order the work, not to close a finding — a workload on an unaffected CPU today can land on vulnerable hardware after an instance resize or a move to newer machines.
Keycloak, Jira, Jenkins: RCE before authentication
Calif demonstrated remote code execution against a list of products in their default configuration. The pre-authentication list is where the business risk concentrates: Keycloak and PingFederate, the identity providers that issue the tokens every other system verifies; Jira, Confluence, Artifactory, and Jenkins, the collaboration and build consoles that sign and ship your software. Compromising any one of those links opens the downstream systems that trust its identities, tokens, or artifacts.
The authenticated list is longer — Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, Netflix Spinnaker — and the researchers say neither list is exhaustive. They also reached RCE on sample applications built with Apache Struts and Spring Boot. Java clients are not spared: the researchers report a client-side exploit in which a Java program is compromised simply by connecting over TLS to a malicious server.
At publication time, no exploitation in the wild had been publicly confirmed. Do not plan around that: the trigger is a short malformed Base64 string, the technique is public, and the researchers say details of the bug leaked before the patch. There is no universal exploit — each product needs its own chain — which slows mass exploitation, but does not protect a specific target such as your identity provider.
The fix and the workaround
The only complete fix is a Java runtime update. Oracle shipped fixed builds on November 17, 2026: 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1, and 8u513-perf for the Java SE 8 Enterprise Performance Pack. Other vendors publish on their own schedules — Amazon already shipped Corretto 17.0.21.1.2 the same day.
If you cannot patch today, a startup flag closes the exposure. The preferred option disables only the vulnerable intrinsic; the alternative turns off a wider set of optimizations, at a broader performance cost.
# preferred: disable only the accelerated Base64 decoder
java -XX:-UseBASE64Intrinsics -jar app.jar
# alternative (Intel/AMD only): turn off AVX optimizations, including the vulnerable one
java -XX:UseAVX=2 -jar app.jar Two precautions. Restart every Java process: a patch or a flag only takes effect when the JVM restarts — verify the running process, not just the installed package. And do not rely on a reverse proxy: request buffering breaks some exploit chains but is not a general mitigation. Applications that decode large volumes of Base64 may slow down with the flag, so test before rolling it out broadly.
Where to start
CISA recommends working in this order, with an owner and a deadline for every system waiting on a vendor build. First, the exposed identity providers — Keycloak, PingFederate. Then exposed Jira, Confluence, Jenkins, and Artifactory consoles. Then other exposed Java services that handle logins, tokens, or file uploads. Finally, internal systems, starting with Kafka, Solr, Elasticsearch, Gerrit, and Spinnaker.
While you wait for a patch, shrink the exposure: take identity, build, and collaboration consoles off the open internet, or put them behind a VPN. CISA also suggests validating Base64 input and enforcing size limits at the application boundary — but size limits alone will not stop this bug, because the malformed input is very short. The JVM flag is the direct stopgap, since it removes the vulnerable code.
Finally, watch for the signals: unexpected JVM crashes or restarts on exposed Java services, Java processes that spawn a shell or open unusual outbound connections, and Java runtimes on hosts your inventory does not account for.
The verdict
If you run an exposed Java identity provider, collaboration console, or build server, apply the fix today — or set -XX:-UseBASE64Intrinsics immediately while you wait for your vendor’s build. The risk concentrates on systems that decode Base64 before authentication, exactly where Calif demonstrated RCE: the fix costs a restart, the compromise costs the trust of everything that depends on your identities.
If your fleet is mostly ARM64 or CPUs without AVX-512 VBMI, the scope is smaller, but do not freeze it — a move to recent hardware or an instance resize can flip you onto the vulnerable side. And check the runtime each application actually runs: a JDK bundled inside a container image does not follow system updates. Java4Shell is a dated, sourced reminder that when the flaw is in the runtime, the installed-package inventory is no longer enough.
References
- Calif — “Java4Shell: Write once, RCE everywhere” (November 17, 2026)
- Qualys — “How to Find, Mitigate, and Patch the Critical Java4Shell Flaw (CVE-2026-114938) Now” (November 17, 2026)
- Akamai — “CVE-2026-114938: Critical Memory Corruption Vulnerability in OpenJDK” (November 17, 2026)
- CISA — “Critical Java Vulnerability Requires Immediate Patching” (November 17, 2026)
- Oracle — Critical Security Patch Update, November 2026