Oracle releases patches for OpenJDK’s Java4Shell flaw


Oracle released patches for Java4Shell on 17 November. Calif says the flaw leaked and its researchers achieved remote code execution in tests. The firm rated it 9.2 and Oracle rated it 8.1 on different severity scales; Oracle had planned fixes for 11 Java vulnerabilities.
Oracle released patches for Java4Shell on 17 November as part of its November Critical Security Patch Update after security firm Calif said it discovered and reported the flaw, which had leaked. The vulnerability, tracked as CVE-2026-114938, affects OpenJDK, an open-source implementation of the Java platform used to run applications. OpenJDK is the upstream source code base for the Java SE reference implementation, and Oracle JDK is built from its source code.
Calif researchers said they had run code remotely against several widely used Java applications in tests, although there is no universal exploit. Earlier, Oracle had scheduled fixes for 11 Java SE vulnerabilities that it said could be exploited over a network without authentication; that does not mean all allow remote code execution. Oracle’s advance notice put the maximum score for the 11 vulnerabilities at 8.1 under CVSS 3.1.
Calif rated Java4Shell critical at 9.2 using version 4.0 of the CVSS severity scale. Oracle rated it high at 8.1 using version 3.1.
Oracle began monthly security releases on 28 May between its quarterly Critical Patch Updates. A Java security plan updated on 10 September scheduled the next monthly release for 17 November.
Oracle’s November notice covered Oracle Java SE, Oracle GraalVM for JDK, Oracle GraalVM Enterprise Edition and Oracle JavaFX. That broader update scope was not a list of products affected by Java4Shell.
The flaw lies in a hardware-accelerated part of Java’s Base64 decoder, which converts encoded data into a usable form. Calif said the resulting memory corruption could let attackers obtain sensitive information or run code. The US Cybersecurity and Infrastructure Security Agency, known as , said specially crafted input could also crash the Java virtual machine, the software that runs Java applications, and allow code execution without authentication.

Researchers identified affected versions starting with OpenJDK 17.0.6, subject to the hardware and decoder in use. Upwind said the vulnerable code runs only on x86-64 processors with AVX-512 VBMI support; not every Java application is exposed. The input must reach the basic or URL decoder in java.util.Base64, not the MIME decoder, after the decoder has been JIT-compiled.
Upwind said it reproduced a 62-byte out-of-bounds write on an AWS c6i instance with Intel Ice Lake hardware. Its reproducer used an oversized buffer to make the write visible without corrupting the heap. The overflow appeared in Temurin 17.0.6, 17.0.21, 21.0.13 and 25.0.5; Java 11.0.33 and 17.0.5 tested clean. Its assessment of Java 27 was based on reading the code, not those tests.
Upwind identified paths to the decoder before authentication through Basic authentication in Spring Security, Tomcat and Jetty and through Auth0’s java-jwt library. It cautioned that reaching the bug does not itself establish damage.
Oracle’s Java 8 Enterprise Performance Pack release notes say it added an AVX-512 Base64 decoding optimization. The researchers had not determined whether that pack was affected.
In its earlier tests, Upwind said a typical token corrupted nothing and it had not demonstrated code execution. Equixly assessed a virtual-machine crash as the realistic outcome. Calif subsequently demonstrated code execution against applications in its own tests, BleepingComputer reported. Calif published a demonstration of the memory corruption but had not released the exploits it used against enterprise applications.
As of 18 November, there was no evidence that the flaw had been exploited in the wild.
Earlier, Equixly reported no reports of exploitation in the wild as of 17 November, and Upwind said the flaw was not in CISA’s Known Exploited Vulnerabilities catalog. Those assessments did not establish that exploitation had not occurred.
At the time of an earlier review, the OpenJDK Vulnerability Group, which coordinates security disclosures affecting OpenJDK source code, had published no advisory newer than 24 October, The Hacker News reported. The Hacker News had previously described the reported 9.2 score and remote-code-execution claims as unconfirmed. Security company watchTowr acknowledged reports of the flaw but cautioned that technical details were still unconfirmed.
OpenJDK has also released patches. CISA urged organizations to identify affected Java installations and apply relevant vendor patches as soon as they are available, prioritizing systems accessible from the internet. Java runtimes can be bundled with other applications, container images or appliances and may not appear separately in software inventories.
In an update reflecting its knowledge as of 20:00 UTC on 17 November, Upwind said Oracle’s fixed builds were available and identified them as 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. At that time, it said fixes for Temurin, Corretto and Zulu were pending. Oracle’s notice covers its own products and does not establish the status of third-party OpenJDK distributions.
Equixly advises prioritizing internet-facing services that decode Base64 from clients. It also advises checking each Java distributor’s advisory for its fixed build and restarting Java processes after updating: replacing files alone does not update a process that is already running.
Where immediate patching is not feasible, CISA recommends temporary controls including checking and restricting Base64 input, limiting its size and limiting access without authentication. It also recommends monitoring for suspicious activity, including unexpected virtual-machine crashes and suspicious Java processes.
Upwind said the HotSpot options -XX:-UseBASE64Intrinsics and -XX:UseAVX=2 stopped the overflow on every affected version it tested. Equixly said neither had vendor confirmation for this CVE and advised testing before rollout. Upwind recommends disabling the Base64 intrinsic where patching is not yet possible; this requires a restart and slows Base64 decoding. Equixly said limiting AVX can also slow cryptography and other vectorized work.
Previdian logged nine exploitation attempts from six IP addresses, while WatchTowr saw reconnaissance and blind attempts; the observations do not establish that a real target was compromised.
4 min