Java4Shell: the OpenJDK flaw that depends on the processor underneath
CVE-2026-114938 lets malformed Base64 input corrupt the heap of Java 17+ runtimes on AVX-512 VBMI processors. Fixes are uneven across distributions, scores disagree, and bundled runtimes make inventory the hard part.
Java4Shell, tracked as CVE-2026-114938, is a pre-authentication memory-corruption flaw in the OpenJDK Base64 decoder, and whether a given server is exposed depends on the processor it runs on as much as on the Java version it runs [1][4]. The flaw affects OpenJDK 17.0.6 and later on x86-64 processors with AVX-512 VBMI support. ARM64 hosts are not affected [1][4]. Oracle and the OpenJDK project published fixes on 17 November 2026 [1][2]. As of 17 and 18 November, Tenable, Cisco Talos and Canada’s Cyber Centre reported no in-the-wild exploitation [5][6][8]. That leaves defenders a short window. It will not stay open for long.
What the advisories say
The defect sits in a hardware-accelerated decoding routine used on processors that support the AVX-512 VBMI instruction set extension. Endor Labs and Datadog Security Labs describe it as a heap overflow: malformed Base64 input causes the JVM to write 62 or 63 bytes past the end of the output buffer before the input is rejected [4][11]. In most cases that crashes the JVM. Calif, which found and reported the flaw, says that combined with application-specific memory behaviour it can leak heap contents or lead to remote code execution [1].
Calif reports pre-authentication code execution in the default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate. It reports post-authentication code execution in Apache Kafka, Apache Solr, Elasticsearch, Gerrit and Spinnaker, and says both lists are not exhaustive [1]. It also demonstrated a client-side case in which a Java program opening a TLS connection to a hostile server can be compromised [1]. Calif says it is withholding the application-specific exploit chains until users have had time to upgrade [1].
The affected processor families include Intel Xeon from Ice Lake onward and AMD Zen 4 and later [1][8]. Older x86-64 parts without VBMI, such as Skylake-SP and Cascade Lake, are not affected. Java 8 and 11 builds and OpenJDK 17.0.0 to 17.0.5 are also outside scope [4]. Oracle’s list adds one exception: Java SE 8 Enterprise Performance Pack build 8u511b33-perf is affected [2][10].
Severity scores do not agree
The published scores differ, and practitioners should note why. Calif rates the flaw 9.2 under CVSS 4.0 [1]. Oracle rates it 8.1 under CVSS 3.1, with network attack vector, no privileges, no user interaction and high attack complexity [2]. The Hacker News reports a preliminary 9.8 from Red Hat and 8.8 from Amazon [8]. TuxCare reports that Red Hat rates the impact Critical without giving a number [9]. Endor Labs notes that Calif has published no vector string for its 9.2 and that it reflects the worst case [4]. The 8.1 figure reflects the effort needed to turn a crash into code execution. It does not mean the flaw is minor.
Assessments of exploitability also differ. VulnCheck told SecurityWeek it does not anticipate that the flaw will be easily or widely exploitable, while Talos said it expects probing attempts to pick up soon after disclosure [7]. Other labs have reproduced crashes and memory disclosure but, according to The Hacker News, not code execution [8]. Calif has published a proof of concept that corrupts memory without demonstrating code execution [5][7]. We read both positions as correct about different attackers. Mass exploitation looks hard. Targeted exploitation of a known product stack looks feasible for a capable team.
Patch availability is uneven
Oracle’s November Critical Security Patch Update contains 11 Java SE fixes and lists Java SE, GraalVM for JDK and GraalVM Enterprise Edition as affected [2]. Tenable and Qualys list the fixed Oracle builds as 8u513-perf, 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1 [5][10]. Downstream builds are arriving on separate schedules. TuxCare’s status table, as of 18 November, showed Amazon Corretto and Azul Zulu fixed, Red Hat pending, Debian fixed only in unstable, and Microsoft, SapMachine and Liberica without confirmed fixed builds [9]. Adoptium has published fixed Temurin versions 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1, with container images following the binaries [12].
Endor Labs warns against inferring a fix from a matching version number, because an LTS build with the same number can still contain the vulnerable code [4]. It reported that the fix had not reached the public OpenJDK 17, 21 and 25 update repositories on 17 November [4].
Appliance vendors are now working through their own products. Cisco’s interim advisory, first published on 18 November, lists ten affected products, including Catalyst Center, Secure Network Analytics and Intersight Virtual Appliance, with fixed releases still to be determined. It lists ISE, IOS XE, IOS XR and NX-OS as under investigation [13]. Expect similar notices from other vendors that ship a Java runtime inside an appliance.
What it means
Java4Shell is mainly an inventory problem. Many products bundle their own runtime, so the system Java version does not show what is actually running [4][10]. Container base images carry their own JDK and must be rebuilt [11]. The hardware condition also changes over time. A workload that is safe today on an older instance can become exposed when it is moved to a newer instance type with VBMI support. No public internet-wide exposure count was available at the time of writing. Censys’s advisory lists affected distributions and products but no host counts [14].
For Malaysia and the wider region, the Cyber Security Agency of Singapore issued an alert on 18 November urging administrators to update promptly [3]. The alert cites Oracle’s version list and the products Calif tested [3]. Malaysian organisations running Keycloak, PingFederate, Jira, Confluence or Jenkins on recent Intel or AMD server hardware, on premises or in regional cloud zones, should treat those hosts as in scope until checked.
What to do
- Patch first. Apply the vendor’s fixed build to every runtime, including those bundled with applications, then restart the Java processes [11][12].
- Prioritise internet-facing identity providers, then internet-facing Jira, Confluence, Jenkins and Artifactory, then other exposed Java services, then internal systems [10].
- Check hardware. Confirm whether each x86-64 host reports AVX-512 VBMI support. Hosts without it, and ARM64 hosts, do not reach the vulnerable code path [1][11].
- Check the running JVM, not the system default. Query the version and flags of each live Java process, run as the same user [4][11].
- Where patching must wait, disable the Base64 intrinsic with the JVM option -XX:-UseBASE64Intrinsics, which Adoptium and Calif document [1][12]. Endor Labs lists a narrower diagnostic option and a broader AVX-512 restriction as alternatives [4]. The AVX restriction will stop an ARM64 JVM from starting, so apply it to x86-64 hosts only [1][11]. Neither option appears in Oracle’s advisory, and neither replaces the patch [5][10].
- Deploy available network detection. Talos has published Snort rules that flag malformed Base64 in specific HTTP request fields [6].
- Track appliance advisories, starting with Cisco’s, and give each system waiting on a vendor build an owner and a deadline [10][13].
Outlook
Calif says the bug took about an hour to find and two more hours to produce a first working exploit [1]. Qualys reports this was done with AI assistance at an AI token cost below 200 US dollars [10]. That gap between discovery and exploit should shape patch planning for runtime-level flaws. Defenders should plan for application-specific exploit chains to appear, from Calif’s delayed release or from others who rebuild them. The work that matters this week is finding every embedded runtime on VBMI-capable hardware before that happens.
Sources
Every R3KONX article cites its primary material. 14 sources, in order of first citation. Links open the original publication.
- Java4Shell: remote code execution in OpenJDK (CVE-2026-114938) Calif · 2026-11-17
- Oracle Critical Security Patch Update Advisory - November 2026 Oracle · 2026-11-17
- High-Severity Vulnerability in Java Products (AL-2026-166) Cyber Security Agency of Singapore · 2026-11-18
- Java4Shell (CVE-2026-114938): Pre-Auth Heap Overflow in OpenJDK Endor Labs · 2026-11-17
- Java4Shell (CVE-2026-114938): Frequently asked questions about OpenJDK remote code execution vulnerability Tenable · 2026-11-17
- CVE-2026-114938: Critical vulnerability in OpenJDK Java Cisco Talos · 2026-11-17
- Java4Shell Vulnerability Puts Many Enterprise Applications at Risk of Takeover SecurityWeek · 2026-11-18
- Java4Shell Flaw (CVE-2026-114938) Enables Pre-Auth RCE in Jira, Jenkins, Keycloak and More The Hacker News · 2026-11-18
- Java4Shell Can Turn One Malformed Base64 String Into RCE on Java 17 Servers TuxCare · 2026-11-18
- How to Find, Mitigate, and Patch the Critical Java4Shell Flaw (CVE-2026-114938) Now Qualys · 2026-11-17
- Java4Shell: Remote Code Execution in OpenJDK (CVE-2026-114938) Datadog Security Labs · 2026-11-17
- Eclipse Temurin guidance for CVE-2026-114938 "java4shell" Eclipse Adoptium · 2026-11
- Vulnerability in Oracle OpenJDK (Java4Shell) Affecting Cisco Products: November 2026 Cisco · 2026-11-18
- Nov 17 Advisory: OpenJDK Remote Code Execution (Java4Shell) [CVE-2026-114938] Censys · 2026-11-17
Researched and written by the R3KONX Analysis Desk from the cited primary material. Exploitation status and downstream patch availability were changing daily at the time of writing, and the OpenJDK advisory and CISA alert could not be retrieved directly. Corrections: event@r3konx.asia.
