On November 17, 2026, the research firm Calif said it had found and reported a vulnerability in Java that it calls Java4Shell, tracked as CVE-2026-114938. Calif made the statement because word of the flaw had leaked before a fix was out, and said it would publish technical details once Oracle released a patch. Oracle shipped its November Java updates the same day. Early reports describe an out-of-bounds write in the Java runtime itself, reachable over the network without authentication, in OpenJDK 17.0.6 and later Java 17 updates, with a CVSS 4.0 score of 9.2. The flaw is in the JDK, not in a library, so dependency scanners that read application packages will not find it. Much of this is still unconfirmed. The CVE record is reserved with no description, Oracle has not tied any of its November fixes to this CVE, and the OpenJDK Vulnerability Group has not published an advisory. This page separates what is confirmed from what is reported, and will be updated as Oracle, OpenJDK and Calif publish. What is confirmed and what is reported Point Status Calif discovered and reported CVE-2026-114938 and named it Java4Shell Confirmed by Calif Oracle’s November 2026 update contains 11 Java SE fixes, all remotely exploitable without authentication, highest CVSS 3.1 score 8.1 Confirmed by Oracle Oracle released JDK 17.0.21.1 on November 17, with 11.0.33.1 and 8u513 as the new security baselines Confirmed by Oracle release notes One of Oracle’s 11 fixes is CVE-2026-114938 Not confirmed. Oracle’s notice lists no CVE IDs Affected range is OpenJDK 17.0.6 and later Java 17 updates Reported by several research teams. One reports Java 17 and above CVSS 4.0 score of 9.2 Reported, attributed to an advisory from South Korea’s National Cyber Security Center The flaw is in the AVX-512 Base64 decoder Reported. Not confirmed by Oracle, OpenJDK or Calif Exploitation in the wild, or a public exploit None reported What the reports say about the flaw The most specific public description places the bug in the JDK’s hardware-accelerated Base64 decoder, the version used on x86 processors with AVX-512. An integer underflow in how the decoder handles padding is said to let a short crafted input write past the end of a buffer on the Java heap. Datadog Security Labs, which has confirmed only the outline, notes that the write happens before the application has a chance to reject the input. That description fits the version boundary. The AVX-512 Base64 decoder was written for JDK 18 and backported to Java 17 in 17.0.6 (OpenJDK bug JDK-8296735, resolved November 9, 2022). The same code had a data-loss bug in December 2023 (JDK-8321599), which appeared only on processors with AVX-512 and disappeared when the JVM was limited to AVX2. That earlier bug produced wrong output, not memory corruption, and has since been fixed. It shows where the code path runs, not that it is the cause here. If the Base64 report is right, an exploitable service needs all of the following: a vulnerable Java runtime an x86 processor with the AVX-512 BW and VBMI features, which means Intel Ice Lake and later server processors and AMD Zen 4 and later, including most current cloud instance types attacker-controlled Base64 reaching the decoder, which is common before authentication: HTTP Basic credentials, JSON Web Tokens, SAML messages and encoded API fields are all decoded before a request is trusted ARM servers, including AWS Graviton and Apple silicon, would not be affected by an x86-only code path. Who is exposed, in priority order Internet-facing Java 17 services on x86 hosts: API gateways, identity and single sign-on services, application servers and anything that accepts tokens or Basic authentication from the internet. Products that bundle their own Java 17 runtime. Many commercial and open-source products ship a JDK inside the install directory or the container image. The operating system’s package manager does not update these, so they wait for the product vendor. Container images built on a Java 17 base image, such as Eclipse Temurin, Amazon Corretto, Azul Zulu, Microsoft Build of OpenJDK and Red Hat images. A running container keeps the old runtime until it is rebuilt and redeployed. Internal Java 17 services that process input from users, partners or message queues. Java 21, 25 and 27. Most reports name only Java 17, but one research team reports Java 17 and above, and Oracle released new builds for every line. Treat these as unconfirmed, not as safe. Java 8 and 11 do not appear in any report of this CVE. Oracle’s update still contains other remotely exploitable fixes for them. Affected and fixed versions Oracle’s November 2026 update lists these Java builds as affected by its 11 fixes, and these replacements: Java line Affected Oracle build Oracle build released November 17 27 27.0.1 27.0.1.1 25 25.0.5 25.0.5.1 21 21.0.13 21.0.13.1 17 17.0.21 17.0.21.1 11 11.0.33 11.0.33.1 8 8u511 8u513 Oracle GraalVM for JDK 17.0.21 and 21.0.13, GraalVM Enterprise Edition 21.3.20.0.3 and JavaFX are on the same affected list. Eclipse Temurin, Amazon Corretto, Azul Zulu, Microsoft, Red Hat and the Linux distributions release on their own schedules and use their own build numbers. None had published a build that names CVE-2026-114938 when this page was written. Check your distribution’s advisory before assuming a build is fixed. How to check a host These commands only read state. Run them on each server or inside each container that runs Java: java -version 2>&1 grep -o -w -E 'avx512bw|avx512_vbmi' /proc/cpuinfo | sort -u java -XX:+PrintFlagsFinal -version 2>/dev/null | grep -E ' (UseAVX|UseBASE64Intrinsics) ' The host matches the reported conditions if the version is 17.0.6 or later and below your vendor’s fixed build, both CPU flags print, UseAVX is 3 and UseBASE64Intrinsics is true. The java on the path is often not the one a service uses. List the running JVMs and the runtime each one loaded: jcmd -l jcmd <pid> VM.version jcmd <pid> VM.flags | tr ' ' '\n' | grep -E 'UseAVX|BASE64' Then find runtimes that packages do not manage: find / -xdev -type f -name release -path '*j*' 2>/dev/null | xargs grep -H '^JAVA_VERSION=' 2>/dev/null On Windows, check java -version for each installed runtime and look for bin\java.exe under application install directories. Mitigation and remediation Install your Java vendor’s November update. For Oracle JDK that is 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1, 11.0.33.1 or 8u513. The update is worth installing whatever it turns out to contain: Oracle says all 11 fixes can be exploited remotely without credentials. Restart every JVM after updating. A running process keeps the old runtime in memory until it is restarted. Rebuild and redeploy container images on an updated base image. Pulling a new base image does not change containers that are already running. Ask vendors of products that bundle Java 17 when they will ship an updated runtime, and whether the bundled JDK can be replaced in the meantime. Consider turning off the Base64 intrinsic where you cannot update yet. Several projects have added this JVM option as a precaution: export JAVA_TOOL_OPTIONS="-XX:-UseBASE64Intrinsics" It makes Java use its plain decoder in place of the processor-specific one, at a small cost in Base64 speed. It addresses the flaw only if the Base64 report is correct, and neither Oracle nor OpenJDK has confirmed it as a workaround. Test it before production use, restart the service, and confirm it took effect with jcmd <pid> VM.flags. It is not a substitute for the update. Reduce exposure. Take Java services off the internet where they do not need to be there, and put administrative interfaces behind a VPN or an allow list. Free SOC trialWant this watched across your environment without building the hunts yourself? Our 24x7 SOC monitors your endpoints, servers and cloud during a free trial.Start a free trialThreat hunting No indicators of compromise have been published, and nobody has reported an attack. Hunting therefore relies on behavior: where Java runs, which services face the internet, JVM crashes, and what Java processes do after they are compromised. Memory corruption exploits often crash the target a few times before they work, so unexplained JVM crashes on internet-facing services are worth a look this week. Example KQL for Microsoft Defender and Microsoft Sentinel The KQL below is example code for Microsoft Defender for Endpoint advanced hunting and Microsoft Sentinel. Adapt table and field names to your environment and test each query before relying on it. Hunt 1: where Java 17 runs // Example: Defender vulnerability management inventory, Java 17 runtimes by version DeviceTvmSoftwareInventory | where SoftwareName has_any ("jdk", "jre", "openjdk", "java", "temurin", "corretto", "zulu", "graalvm") | where SoftwareVersion startswith "17." | summarize Devices=dcount(DeviceId), DeviceList=make_set(DeviceName, 50) by SoftwareVendor, SoftwareName, SoftwareVersion | order by Devices desc The inventory misses runtimes bundled inside applications. This finds them by the path of the running binary: // Example: Defender for Endpoint, every java binary that has run, with its path DeviceProcessEvents | where Timestamp > ago(14d) | where FileName in~ ("java", "java.exe", "javaw.exe") | summarize Starts=count(), LastSeen=max(Timestamp), SampleCommand=any(ProcessCommandLine) by DeviceName, FolderPath, ProcessVersionInfoProductVersion | order by DeviceName asc Hunt 2: Java services that accept connections from the internet // Example: Defender for Endpoint, Java processes accepting inbound connections from public addresses DeviceNetworkEvents | where Timestamp > ago(14d) | where ActionType == "InboundConnectionAccepted" | where InitiatingProcessFileName in~ ("java", "java.exe", "javaw.exe") | where RemoteIPType == "Public" | summarize Connections=count(), Sources=dcount(RemoteIP), Ports=make_set(LocalPort, 20), LastSeen=max(Timestamp) by DeviceName, InitiatingProcessFolderPath | order by Sources desc Patch and check these hosts first. Hunt 3: Java starting a shell or a download tool // Example: Defender for Endpoint, shells, script hosts and download tools started by a Java process DeviceProcessEvents | where Timestamp > ago(14d) | where InitiatingProcessFileName in~ ("java", "java.exe", "javaw.exe") | where FileName in~ ("sh", "bash", "dash", "zsh", "cmd.exe", "powershell.exe", "pwsh.exe", "python", "python3", "perl", "curl", "wget", "nc", "ncat", "certutil.exe", "mshta.exe", "bitsadmin.exe", "rundll32.exe") | project Timestamp, DeviceName, AccountName, InitiatingProcessFolderPath, InitiatingProcessCommandLine, FileName, ProcessCommandLine | order by Timestamp desc Build tools, application servers and schedulers start shells legitimately. Establish what is normal for each service, then alert on the rest. Hunt 4: JVM crashes A crashed JVM writes an hs_err_pid<number>.log file to its working directory. // Example: Defender for Endpoint, JVM fatal error logs created on servers DeviceFileEvents | where Timestamp > ago(14d) | where ActionType == "FileCreated" and FileName startswith "hs_err_pid" | summarize Crashes=count(), FirstSeen=min(Timestamp), LastSeen=max(Timestamp), Folders=make_set(FolderPath, 10) by DeviceName | order by Crashes desc For Linux services whose output goes to the system log: // Example: Sentinel Syslog, JVM fatal errors and segmentation faults in Java services Syslog | where TimeGenerated > ago(14d) | where SyslogMessage has "A fatal error has been detected by the Java Runtime Environment" or (SyslogMessage has "SIGSEGV" and SyslogMessage has_any ("libjvm", "java")) | summarize Crashes=count(), FirstSeen=min(TimeGenerated), LastSeen=max(TimeGenerated), Sample=any(SyslogMessage) by Computer | order by Crashes desc Open the crash log for any internet-facing service. A crash in a Base64 or StubRoutines frame matches the reported flaw and should be treated as a possible exploitation attempt. Hunt 5: new outbound connections from Java services // Example: Defender for Endpoint, public destinations a Java service contacted this week but not in the prior 30 days let Baseline = DeviceNetworkEvents | where Timestamp between (ago(37d) .. ago(7d)) | where InitiatingProcessFileName in~ ("java", "java.exe", "javaw.exe") | where ActionType == "ConnectionSuccess" and RemoteIPType == "Public" | distinct DeviceName, RemoteIP; DeviceNetworkEvents | where Timestamp > ago(7d) | where InitiatingProcessFileName in~ ("java", "java.exe", "javaw.exe") | where ActionType == "ConnectionSuccess" and RemoteIPType == "Public" | join kind=leftanti Baseline on DeviceName, RemoteIP | summarize Connections=count(), FirstSeen=min(Timestamp), Ports=make_set(RemotePort, 10) by DeviceName, RemoteIP, RemoteUrl, InitiatingProcessFolderPath | order by FirstSeen desc Responding to a suspected compromise Keep the evidence before restarting anything: the hs_err_pid crash logs, the application and access logs, and a memory image of the Java process if you can take one. Isolate the host from untrusted networks, then update the Java runtime and restart the service. Review what the service account did after the first crash or suspicious request: child processes, new files in the application and temporary directories, scheduled tasks or cron entries, and outbound connections. Rotate the secrets the service could read: database passwords, API keys, signing keys, cloud credentials and anything in its configuration or environment. Rebuild the host or container if code execution is confirmed or cannot be ruled out. Timeline November 9, 2022: The AVX-512 Base64 decoder is backported to Java 17 for release 17.0.6 (JDK-8296735). December 8, 2023: A data-loss bug in the same decoder is reported (JDK-8321599) and later fixed. It is a correctness bug, not a security flaw. November 16, 2026: Reports cite an advisory attributed to South Korea’s National Cyber Security Center that rates CVE-2026-114938 at 9.2 under CVSS 4.0. November 17, 2026: Calif states that it discovered and reported Java4Shell, that the flaw leaked, and that it will publish details after Oracle’s patch. November 17, 2026: Oracle releases its November 2026 Critical Security Patch Update with 11 Java SE fixes and new JDK builds for Java 8, 11, 17, 21, 25 and 27. November 17, 2026: Datadog Security Labs and Rapid7 publish preliminary notes. Rapid7 ships an authenticated check for its scanners. November 18, 2026: The CVE record is still reserved. OpenJDK has not published an advisory, and Calif’s technical write-up has not appeared. Frequently Asked Questions Is Java4Shell as bad as Log4Shell? Too early to say. Log4Shell was trivial to exploit and public exploit code appeared within hours. Java4Shell is reported as a memory corruption flaw, which is usually harder to turn into reliable code execution, and it appears to depend on the processor. What it has in common is reach: the flaw is in the runtime, so any Java 17 application could be exposed. Does updating my application’s libraries fix it? No. The flaw is reported in the JDK itself. Update the Java runtime the application runs on, including any runtime bundled with a product or built into a container image. Is Java 8, 11 or 21 affected? No report names Java 8 or 11 for this CVE. Java 21 and later are unconfirmed: most reports name only Java 17, one says Java 17 and above. Oracle released updates for every supported line on November 17, and all of them contain fixes for remotely exploitable flaws. We use Temurin, Corretto or a Linux distribution’s OpenJDK. What do we install? Your distribution’s own update, when it is released. Oracle’s build numbers, such as 17.0.21.1, apply to Oracle JDK only. Until your vendor publishes, consider the JVM option above and reduce exposure. Does the -XX:-UseBASE64Intrinsics option fix it? Only if the public reports about the Base64 decoder are right, which Oracle and OpenJDK have not confirmed. It is a reasonable precaution with little cost, not a replacement for the update. Will my software composition analysis tool find this? Probably not. Those tools read application dependencies such as jar files. The JDK is a runtime, so look for it in host and container inventories. Is CVE-2026-114938 being exploited? No exploitation and no public exploit had been reported by November 18, 2026, and it is not in the CISA KEV catalog. That can change quickly once technical details are published. How ThreatDefence is responding ThreatDefence has added CVE-2026-114938 to our vulnerability intelligence. Across all monitored environments, our SOC is: identifying servers and containers that run Java 17, starting with internet-facing services watching Java processes for crashes, unexpected child processes and new outbound connections helping customers confirm their Java vendor’s update and the restart of each service If you want help finding Java runtimes across your environment or investigating a suspicious crash, contact us. Read more about our managed detection and response. Sources Oracle, Critical Security Patch Update, November 2026 (November 17, 2026) Oracle, JDK 17.0.21.1 release notes (November 17, 2026) Calif, statement on Java4Shell and CVE-2026-114938 (November 17, 2026) CVE Program, CVE-2026-114938 record (reserved, checked November 18, 2026) OpenJDK Vulnerability Group, advisories (checked November 18, 2026) Datadog Security Labs, Java4Shell: remote code execution in OpenJDK (November 17, 2026) Rapid7, CVE-2026-114938: reported critical pre-authentication remote code execution in Java 17 (November 17, 2026) Oligo Security, Java4Shell (November 17, 2026) OpenJDK, JDK-8296735: Base64 decoding optimization for x86 using AVX-512, backport to 17.0.6 OpenJDK, JDK-8321599: Data loss in AVX3 Base64 decoding Elastic, detection-rules issue 7199: suspicious Java child process execution (November 17, 2026)