Nov 17 Advisory: OpenJDK Remote Code Execution (Java4Shell) [CVE-2026-114938]

Rapid Response

This is a developing issue. Details are still emerging, and we may update this advisory as more information becomes available.

Vulnerability Description

CVE-2026-114938, nicknamed Java4Shell, is a memory corruption vulnerability in OpenJDK that Calif reports can be exploited remotely, in many cases without authentication, to execute code. It affects Java running on x86-64 processors with AVX-512 VBMI, starting with OpenJDK 17.0.6. Amazon Corretto has released fixed builds for Java 17, 21, 25, and 27, and Oracle released JDK 17.0.21.1 on November 17, 2026. All Java users should update immediately.

The flaw is in OpenJDK’s x86-64 Base64 decoder, in the hand-rolled assembly path that uses AVX-512 VBMI instructions. The store mask is miscalculated for the padded tail of the input, so an underflowed output length produces a full-width write that runs past the end of the decoded output before the decoder rejects the input. Calif, the firm that discovered and reported the flaw, calls it Java4Shell in its write-up and rates it 9.2 under CVSS v4.0. Oracle, which assigned the CVE, rates it 8.1 under CVSS v3.1 in its CVE record and describes it as a “difficult to exploit vulnerability” in the Compiler component that “allows unauthenticated attacker with network access via multiple protocols to compromise” Java SE and GraalVM, and that “can only be exploited by supplying data to APIs in the specified Component”, for example through a web service. Oracle says successful attacks “can result in takeover” of the affected product.

Exploitation requires an x86-64 processor with AVX-512 VBMI, which includes Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids, and Granite Rapids, and AMD EPYC and Ryzen Zen 4 or newer. ARM64 processors are not affected. On Linux, all four of the avx512f, avx512bw, avx512vbmi, and bmi2 flags must appear in /proc/cpuinfo for a host to have the required instructions. The flaw is reachable by a remote attacker wherever an application passes attacker-controlled input to the decoder, and ordinary features such as authentication and file uploads can do so. A Java program that opens a TLS connection to a malicious server can also be compromised even if it exposes no service of its own. With careful exploitation, the corruption can also expose memory without any code execution, in the style of Heartbleed.

Calif additionally describes corrupting the length of an adjacent array so that more of the heap is copied back into the application’s normal response, which it calls JavaBleed. Calif says this lets an attacker read the contents of the Java heap, which may hold PII, private keys, other secret keys, and tokens, and that it had this capability before it achieved code execution.

Calif’s own testing achieved remote code execution in default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins, and PingFederate without authentication, and of Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Netflix Spinnaker after authentication. The lists are not exhaustive. There is no universal exploit: depending on how an application uses Base64 and how it works, different steps may be required, and Calif is withholding the per-product chains. Calif says it sped up exploit development by first building pipelines and strategies and then directing AI agents to use them for the initial exploitation, after which most of the later chains were built almost autonomously with AI. These results are Calif’s own and have not been independently confirmed.

Impact

An unauthenticated remote attacker can corrupt the heap of a Java process, read from it, and, as Calif’s testing shows, execute code with that application’s privileges given per-application work. For server software those privileges usually include stored credentials, tokens, and data, so the exposure extends to whatever those credentials authenticate to. The absence of a universal exploit does not make an application safe, because Calif says the per-application chains were built largely with AI assistance once the initial strategies existed. Update affected Java installations, including clients that connect to servers you do not control, and restart every JVM afterward so the fixed build is the one running. Where an instance was reachable while unpatched, review it for signs of use and rotate its credentials if compromise cannot be ruled out.

Alongside updating, two JVM options mitigate the flaw: -XX:-UseBASE64Intrinsics disables the Base64 intrinsics, and -XX:UseAVX=2 disables AVX-512 and should be applied only on x86-64 hosts, because it fails to start on ARM. Calif recommends either but gives no data on how effective they are or what it costs in performance.

Details

CVECVE-2026-114938. CVSS v3.1 8.1 (High), assigned by Oracle. Calif’s write-up gives a CVSS v4.0 score of 9.2 (Critical).
Date of DisclosureNovember 17, 2026
Affected AssetsJava applications running an affected OpenJDK build on x86-64 processors with AVX-512 VBMI. Calif lists Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine, Red Hat OpenJDK, Oracle OpenJDK, Oracle Java SE, and Oracle GraalVM for JDK as vulnerable, and Oracle also lists Oracle GraalVM Enterprise Edition. ARM64 is not affected.
Vulnerable Software VersionsOpenJDK from 17.0.6 up to the fixed releases. Oracle’s advisory lists Oracle Java SE 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3, and 27.0.1.0.3, Oracle GraalVM for JDK 17.0.21.0.3 and 21.0.13.0.3, and Oracle GraalVM Enterprise Edition 21.3.20.0.3 as affected. Calif lists the same Java SE builds. Oracle’s list includes only the 8u511b33-perf build of Java 8 and no Java 11 build, so it does not list the standard Java 8 build or Java 11 as affected.
PoC AvailableCalif’s write-up includes a trigger program that reproduces the memory corruption but not code execution, and Calif says it is safe to run because every write lands inside a real array. Calif says it will publish the per-product chains later. Calif also says samczsun shared a client-side exploit with it and does not state whether that exploit is public.
Exploitation StatusCalif reports demonstrating remote code execution in its own testing and says there is no universal exploit, so each application may need different steps. No exploitation in the wild has been reported as of November 17, 2026.
Patch StatusAmazon Corretto 17.0.21.1.2, 21.0.13.1.2, 25.0.5.1.2, and 27.0.1.1.4 list CVE-2026-114938 in their changelogs under the hotspot/compiler component. Oracle JDK 17.0.21.1 (17.0.21.1+4) was released November 17, 2026, and its release notes say it contains fixes for the vulnerabilities described in the Oracle Critical Patch Update without naming CVE-2026-114938. The same notes list 11.0.33.1 and 1.8.0_513-b04 as the current security baselines for Java 11 and Java 8, and do not tie those builds to this CVE. Calif says Oracle’s patch went out at 1:00 p.m. Pacific and that OpenJDK released a security advisory. Oracle’s advisory lists CVE-2026-114938 in its Java SE risk matrix, which covers 11 new Java SE patches, with the affected builds above, and does not name fixed builds. Fixed builds for other Java lines and distributions have not been confirmed here.

References