The Java4Shell vulnerability is a serious memory corruption flaw affecting Java applications on specific x86-64 servers. Disclosed on 17 November 2026, it can let an unauthenticated attacker crash the Java Virtual Machine (JVM) or potentially execute code remotely.
Tracked as CVE-2026-114938, the flaw is particularly significant because vulnerable Java runtimes may be embedded within third-party products. Organisations must therefore check the runtime used by each application, rather than relying only on an inventory of separately installed Java packages.
Java4Shell vulnerability disclosed in Oracle update
Oracle released its November 2026 Critical Security Patch Update on 17 November 2026. The wider update contains 655 security patches, 62 of them rated critical, and affects 62 products. The OpenJDK project published a corresponding vulnerability advisory on the same day.
Oracle gives CVE-2026-114938 a CVSS v3.1 score of 8.1, classified as high severity. Researchers using CVSS v4.0 have assessed it at 9.2, which falls within the critical range. This difference reflects the scoring systems used, rather than disagreement about the underlying technical risk.
Security research company Calif also published its Java4Shell technical analysis and a safe trigger proof of concept on 17 November 2026. Red Hat, Qualys, Datadog and SentinelOne subsequently released analysis or guidance, followed by further public warnings and reporting on 18 November 2026.
Disclosure timeline
- 17 November 2026: Oracle released its November Critical Security Patch Update with corrected Java builds.
- 17 November 2026: OpenJDK published its vulnerability advisory and identified fixed upstream releases.
- 17 November 2026: Calif disclosed the technical cause, exploitation approach and safe trigger proof of concept.
- 17 November 2026: Red Hat published bulletin RHSB-2026-018, including interim mitigations.
- 18 November 2026: Additional guidance and reports appeared, including an advisory from Japan’s IPA.
Which Java versions and systems are affected?
The Java4Shell vulnerability does not affect every Java installation. The vulnerable code path is selected only on x86-64 processors supporting AVX-512 VBMI instructions. Relevant hardware includes Intel processors from Ice Lake through Granite Rapids and AMD Zen 4 or newer. ARM64 systems and x86-64 processors without AVX-512 VBMI are not known to be affected.
Affected OpenJDK lines include version 17.0.6 and later unpatched releases, along with corresponding Java 21, 25 and 27 builds. Oracle specifically lists Java SE 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3 and 27.0.1.0.3 as affected.
Standard OpenJDK 8 and 11 builds are not affected. However, Oracle Java SE 8 Enterprise Performance Pack 8u511b33-perf is affected. This exception is important because early summaries stating that Java 8 was entirely unaffected could lead organisations to overlook that specialist release.
Oracle GraalVM for JDK 17.0.21.0.3 and 21.0.13.0.3 is also listed as affected. Downstream distributions, including Eclipse Temurin, Amazon Corretto, Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine and Red Hat OpenJDK, can inherit the issue until corrected builds are installed.
Fixed Java releases
OpenJDK identifies 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1 as fixed upstream builds. Organisations using Oracle Java, GraalVM or another OpenJDK distribution should follow the relevant supplier’s advisory and install its patched build or a later supported release.
How the Java4Shell vulnerability works
The flaw is inside HotSpot’s accelerated Base64 decoder. On compatible processors, Java uses an AVX-512 VBMI intrinsic to decode Base64 more quickly. Certain malformed endings cause the decoder to calculate the output length incorrectly, resulting in an all-ones store mask.
A low-level evmovdqub operation can then write 64 bytes beyond the end of the destination array. Crucially, this memory corruption happens before Java’s higher-level validation rejects the malformed Base64 input.
On OpenJDK 21 and newer, the published trigger uses 64 letter A characters followed by one equals character. OpenJDK 17 requires 128 letter A characters followed by one equals character to reach the vulnerable accelerated code. The public proof of concept safely demonstrates the out-of-bounds write rather than providing a universal exploit.
An application becomes remotely exposed when it decodes attacker-controlled Base64 data. Potential entry points include HTTP Basic authentication credentials, SAML artefacts, security tokens and some file upload processes. If decoding occurs before authentication, an attacker may reach the flaw without holding a valid account.
Calif reported laboratory demonstrations of pre-authentication remote code execution against Keycloak, Atlassian Jira and Confluence, Jenkins, JFrog Artifactory and PingFederate. Authenticated exploitation chains were demonstrated for Kafka, Solr, Elasticsearch, Gerrit and Spinnaker. This is an indicative list, not confirmation that every deployment of those products is exploitable.
From a crash to remote code execution
The immediate result may be a JVM crash, causing a denial of service. Achieving remote code execution is more complex. Researchers describe using heap grooming to turn the fixed-pattern overflow into controlled memory writes, followed by arbitrary write or information disclosure capabilities.
These chains require work for each target application and can be sensitive to request timing and network conditions. Java4Shell is therefore not a single exploit that works identically against every Java service, but its location in widely used decoding functionality creates substantial exposure.
Current Java4Shell exploitation status
As of 18 November 2026, no confirmed in-the-wild exploitation or attributed threat actor has been publicly reported. The trigger proof of concept and exploitation methodology are public, however, so defenders should not interpret the absence of confirmed attacks as a reason to delay updates.
Oracle and OpenJDK have not published authoritative malicious IP addresses, file hashes or other compromise indicators. SentinelOne suggests searching logs for suspicious Base64 values at least 65 characters long, ending with one or two equals characters, and with a total length modulo 64 of 1, 2 or 3. This is only a detection heuristic and may produce legitimate matches.
Why this Java flaw matters to organisations
Many identity platforms, development tools, collaboration systems and search services include their own Java runtime. Cloud instances may also use AVX-512-capable processors by default. Internet-facing applications that decode Base64 before login should receive the highest priority because they may expose the Java4Shell vulnerability without requiring valid credentials.
What organisations should do now
- Inventory the exact JVM build used by each Java service, including runtimes bundled inside commercial and open-source products.
- Confirm whether the underlying x86-64 processor exposes AVX-512 VBMI instructions.
- Upgrade to 17.0.21.1, 21.0.13.1, 25.0.5.1 or 27.0.1.1, or the equivalent fixed build supplied by the relevant distributor.
- Prioritise internet-facing identity, administration, continuous integration and collaboration platforms that process Base64 input before authentication.
- Review logs for unusual malformed Base64 requests, while treating pattern matches as attempted-trigger heuristics rather than proof of compromise.
If patching cannot be completed immediately, Red Hat recommends starting affected JVMs with -XX:-UseBASE64Intrinsics to disable the vulnerable accelerated Base64 path. An alternative is -XX:UseAVX=2, which disables AVX-512 entirely, but this option must not be used on ARM. These temporary settings should be removed after the appropriate Java update has been installed and tested.
Originally reported by theregister.com.








