Skip to content

Vulnerabilities & Patches

Oracle and OpenJDK patch critical Java4Shell remote code execution flaw

CERT-EU recommends that organizations update vulnerable OpenJDK versions, prioritize internet-facing applications and restart all JVMs.

https://breach-signal.com/vulnerabilities/oracle-openjdk-patch-critical-java4shell-rce-flaw/

Share

Key facts

  • The flaw is a memory corruption vulnerability in OpenJDK's Base64 decoder that can cause denial of service and, in some cases, unauthenticated remote code execution.
  • Exploitation requires an x86-64 CPU with avx512f, avx512bw, avx512vbmi and bmi2, including Intel Xeon Ice Lake or later and AMD Zen 4 or later.
  • Affected OpenJDK-based distributions include Eclipse Temurin, Amazon Corretto, Azul Zulu, Microsoft Build of OpenJDK, BellSoft Liberica, SAP SapMachine and Red Hat build of OpenJDK.
  • Calif demonstrated unauthenticated remote code execution in default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate.

Oracle and the OpenJDK project released fixes for CVE-2026-114938 on Nov. 17, 2026, for affected OpenJDK-based distributions used in internet-facing Java applications.

The flaw, nicknamed Java4Shell, is a memory corruption vulnerability in OpenJDK’s Base64 decoder that can cause denial of service and, in some cases, unauthenticated remote code execution. It lives in handwritten x86-64 assembly that OpenJDK HotSpot uses to accelerate Base64 decoding with AVX-512 VBMI instructions. Oracle rates it 8.1 under CVSS 3.1, while Calif rates it 9.2 Critical under CVSS 4.0.

Affected Oracle Java SE and OpenJDK 17 runs from 17.0.6 to 17.0.21, with fixed release 17.0.21.1. Version 21 includes 21.0.13 and earlier, fixed at 21.0.13.1; version 25 includes 25.0.5 and earlier, fixed at 25.0.5.1; and version 27 includes 27.0.1 and earlier, fixed at 27.0.1.1. Oracle Java SE 8 performance builds include 8u511-perf, with fixed release 8u513, and Oracle GraalVM for JDK 17 and 21 releases are affected. Affected OpenJDK-based distributions include Eclipse Temurin, Amazon Corretto, Azul Zulu, Microsoft Build of OpenJDK, BellSoft Liberica, SAP SapMachine and Red Hat build of OpenJDK.

The vulnerability requires an x86-64 CPU with avx512f, avx512bw, avx512vbmi and bmi2, including Intel Xeon Ice Lake or later and AMD Zen 4 or later. ARM64 systems are not affected, and the Base64 intrinsic is active by default when the CPU supports the required features. Calif demonstrated unauthenticated remote code execution in default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate, plus authenticated remote code execution in Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit and Spinnaker. Oracle notes the flaw can only be exploited by supplying data to the affected APIs, such as through a web service, and cannot be exploited through untrusted Java Web Start applications or applets.

A malformed Base64 input of 64 A characters followed by one equals sign can trigger the corruption, while older OpenJDK 17 versions require 128 A characters. In confirmed malformed-tail cases, the overflow performs a 64-byte store that lands 62 or 63 bytes past the decoded-output length. Turning that overwrite into code execution requires influencing which objects occupy the affected memory, converting the corruption into useful access, and identifying application or runtime behavior that permits further execution. Crashing an affected JVM is trivial, and repeated requests can keep a service down, according to VulnCheck.

Calif also demonstrated remote code execution in sample applications built on Apache Struts and Spring Boot. Calif says there is no universal exploit and each target requires tailored steps including a decoder route, a feedback path, a heap layout and an application-specific action. Attacker-controlled Base64 data must reach the decoder, for example through HTTP Basic authentication, JWT, SAML messages, cookies or TLS data. SentinelOne notes that java.util.Base64.Decoder can carry the payload in any value, such as HTTP Basic credentials, SAML artifacts, CSRF tokens or MIME encoded words.

The California-based security firm Calif found the flaw and published details on Nov. 17 after some information leaked before coordinated disclosure. Researcher Quang Luong discovered it using OpenAI’s Codex AI coding assistant during an audit of OpenJDK internals, spending about one hour on discovery and two hours on an initial exploit at under $200 in AI cost. Calif published a proof of concept that checks whether a given JDK and CPU combination reproduces the memory corruption, but is withholding product-specific exploit chains to give organizations time to upgrade.

As of Nov. 18, 2026, no attacks exploiting the vulnerability had been reported in the wild. Cisco Talos has not observed in-the-wild exploitation but anticipates probing attempts may pick up soon after public disclosure. VulnCheck does not expect widespread exploitation but expects narrow paths to remote code execution in specific applications, and watchTowr likewise does not expect indiscriminate exploitation.

CERT-EU recommends updating vulnerable OpenJDK versions, prioritizing internet-facing applications, and restarting all JVMs after the update. CISA published an emergency advisory on Nov. 17, 2026, urging organizations to apply patches as soon as they become available. Rapid7 urges organizations to identify and remediate affected systems, prioritizing internet-exposed ones on an emergency footing outside normal patching cycles. It also advises verifying that each running Java process uses the patched build and rebuilding affected container images.

Where immediate patching is not possible, administrators can disable the intrinsic with -XX:+UnlockDiagnosticVMOptions -XX:-UseBASE64Intrinsics, or disable AVX-512 with -XX:UseAVX=2; both options require a JVM restart and the AVX-512 option can reduce performance. Cisco Talos provides SNORT rules, including Snort2 SIDs 1:67502 through 1:67513 and Snort3 SIDs 1:301754 and 1:67502 through 1:67513. Rapid7’s Nov. 17 content release 1.1.4115 includes an authenticated vulnerability check available to its scan engine and Rapid7 Agent.

Kevin Beaumont said the flaw is not world ending and organizations should patch as they normally do, according to Help Net Security. Gal Elbaz, co-founder and CTO of Oligo Security, told CyberScoop that technical barriers are surmountable and organizations should treat this as a serious exposure requiring compensating controls.

Calif frames the bug as a byproduct of software complexity because the AVX-512 speedup shipped with every copy of Java though few programs need it. The firm says researchers were “looking for copper and found gold.” It argues that AI can help maintainers identify and remove unnecessary features and optimizations, presenting the same AI-assisted workflow that found the bug as a way to reduce attack surface rather than only to discover and weaponize flaws.

OpenJDK is an open-source implementation of Java SE co-maintained by Oracle and a global community including Red Hat, Amazon and Microsoft, and it runs most Java applications. Oracle says the flaw can be exploited remotely over a network without authentication, carries high attack complexity, and has high impact on confidentiality, integrity and availability. Rapid7 says an unauthenticated remote attacker can crash a Java application or execute code under the Java process’s privileges.

The published OpenJDK fix hands final Base64 groups shorter than four bytes to the ordinary Java decoder instead of the affected optimized routine. Rapid7 says Java rejects the malformed input afterward, but catching the exception does not reverse the corruption, and harm to another object may stay hidden until later activity such as garbage collection crashes the JVM.

Oracle’s November Critical Security Patch Update addresses 655 security vulnerabilities across its product portfolio, while CyberScoop reports the update covered 652 flaws. The advisory contains 11 new Java SE patches and credits Quang Luong of Calif.io and Albert Yang for CVE-2026-114938. Calif said the find was its second major AI-assisted disclosure in recent weeks, after last month’s disclosure of php2shell.

Was this helpful?

Join the discussion

Please read our Comment Policy before commenting. First-time comments are held for moderation. By posting a comment you agree to our Terms of Use and Privacy Policy, which explain how your name, email address and comment are used.

Your email address will not be published. Required fields are marked *