Java4Shell (CVE-2026-114938): Java SE and OpenJDK Fix Guide
Java4Shell (CVE-2026-114938) is a heap overflow in Java 17.0.6 and later on x86-64 CPUs with AVX-512 VBMI. Who is affected, how to check, how to fix it.
Akshay Dubey
Co-Founder
Java4Shell is a memory-corruption bug in the code Java uses to decode Base64 text. It affects Java 17.0.6 and later on x86-64 processors with AVX-512 VBMI, which include many Intel Xeons and AMD Zen 4 or newer, and fixed builds have been out since November 17, 2026. The facts below are as of November 21, 2026, with sources at the end. Our record is CVE-2026-114938 on CVE Pulse.
In short:
- What: a heap buffer overflow in the HotSpot JVM's fast Base64 decoder on x86-64 processors with AVX-512 VBMI. Oracle's official score is 8.1 (High).
- Who: Java 17.0.6 and every later unfixed release, in OpenJDK and HotSpot builds such as Oracle Java SE, plus Oracle Java SE 8 Enterprise Performance Pack builds. Unsupported Java 18 to 20, 22 to 24 and 26 have no fix. Standard Java 8 and 11, and ARM64 machines, are not affected.
- Fix: update to 17.0.21.1, 21.0.13.1, 25.0.5.1 or 27.0.1.1, or your vendor's fixed build. Until then, start Java with
-XX:-UseBASE64Intrinsics. - Attacks: private working exploits and attack attempts are reported, but none of the reports we found confirms a successful attack; it is not in CISA's Known Exploited Vulnerabilities (KEV) catalog.
What is Java4Shell?
Java4Shell is the name the security firm Calif gave to CVE-2026-114938, a bug that lets malformed Base64 input make Java write past the end of a memory buffer. To decode Base64 faster, Java's HotSpot virtual machine uses AVX-512 VBMI instructions on processors that have them, and this fast path mishandles a short padded ending. For 64 letter A's followed by one "=", Java sets aside a 50-byte result but writes 64 bytes, 62 of them past the end (on OpenJDK 17, Calif says, it takes 128 A's). Java then rejects the input, but neighboring memory may already be overwritten.
Base64 carries login tokens, file uploads and messages between services, so such a string can reach a server in many ways. Oracle, which issued the CVE, calls it "difficult to exploit" and says it "allows unauthenticated attacker with network access via multiple protocols" to compromise Java.
Calif says careful exploitation can leak memory, as Heartbleed did, or run the attacker's code. It reports code execution without a login against default setups of Keycloak, Jira, Confluence, JFrog Artifactory, Jenkins and PingFederate, and with a login against Kafka, Solr, Elasticsearch, Gerrit and Spinnaker. Elastic has confirmed the Elasticsearch result: on affected processors, a user with at least "read" access can run code in versions 7.17.3 and later. The other results are Calif's own; it says each target needs its own exploit chain, and it is holding these back for now.
How serious is it?
Oracle's official score is 8.1 (High). Red Hat rates it 9.8: its vector matches Oracle's except for attack complexity, which Red Hat scores Low and Oracle High. Calif's 9.2 uses the newer CVSS 4.0, which does not compare directly with 3.x scores, and Calif has not published its vector.
| Who | Score | Scale | Notes |
|---|---|---|---|
| Oracle (issued the CVE) | 8.1 High | CVSS 3.1 | The official score. AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H |
| NVD and GitHub advisory GHSA-6g7j-4xh3-xq2q | 8.1 High | CVSS 3.1 | Both show Oracle's score and vector |
| Red Hat | 9.8 Critical | CVSS 3.0 | AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H |
| Calif (the researchers) | 9.2 Critical | CVSS 4.0 | Calif's own rating; vector not published |
On November 17, CISA published an alert titled "Critical Java Vulnerability Requires Immediate Patching", with input from Canada's Cyber Centre.
Am I affected?
You are affected only if all three are true: your Java is 17.0.6 or any later release, or an Oracle Java SE 8 Enterprise Performance Pack build, without the fix; it is a HotSpot build of OpenJDK, such as Oracle Java SE, Eclipse Temurin, Amazon Corretto, Azul Zulu or Red Hat's (OpenJ9-based IBM Semeru is not affected, its maintainers say); and it runs on an x86-64 processor with AVX-512 VBMI.
| Java line | Affected | Fixed from |
|---|---|---|
| 8 | No, except Oracle Java SE 8 Enterprise Performance Pack: 8u511b33-perf and, Oracle says, likely earlier builds | Oracle's November 2026 Critical Security Patch Update |
| 11 | No | Not affected by this CVE |
| 17 | From 17.0.6, before 17.0.21.1 | 17.0.21.1 |
| 21 | Before 21.0.13.1 | 21.0.13.1 |
| 25 | Before 25.0.5.1 | 25.0.5.1 |
| 27 | Before 27.0.1.1 | 27.0.1.1 |
| 18 to 20, 22 to 24, 26 | All releases (no longer supported) | None so far: move to a fixed 21, 25 or 27 |
Oracle GraalVM for JDK 17.0.21.0.3 and 21.0.13.0.3, and GraalVM Enterprise Edition 21.3.20.0.3, are also affected.
Processors with AVX-512 VBMI include, according to Calif, Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids and Granite Rapids, and AMD EPYC and Ryzen from Zen 4 on. Calif and Red Hat both say ARM64 is not affected.
The fast decoder runs only once Java's top-tier just-in-time compiler (C2 in OpenJDK) has compiled it, which an attacker can force by first sending valid Base64. Microsoft says the default Azure Functions Java worker uses only the C1 compiler, which does not call the vulnerable code, so that default setup is not exposed.
How do I check?
Check the Java version and the processor everywhere Java runs; you need both answers.
1. List every Java runtime. CISA advises including the ones bundled inside applications, containers and appliances.
2. Check the version. Run java -version with each runtime, inside the container for a container image, and compare the first line, for example openjdk version "25.0.5.1" 2026-11-17, with the table above.
3. Check the processor (Linux). Run Calif's command on the machine or virtual machine where Java runs:
grep -m1 '^flags' /proc/cpuinfo | tr ' ' '\n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'If it prints all four flags, the processor can run the vulnerable code. If any is missing, or uname -m prints aarch64, it cannot.
4. Look up a version. In our free Am I Affected? checker, enter the product openjdk and the version from that first line, such as 17.0.10, not a vendor build number: it lists the CVEs whose records match, this one for Java 17, 21, 25 and 27 only, and errs toward showing old records that name no version. It checks the version, not the processor. The same lookup is in our free CVE API:
curl "https://app.echelongraph.io/api/v1/public/cves/match?product=openjdk&version=17.0.10"How do I fix it?
Update Java to 17.0.21.1, 21.0.13.1, 25.0.5.1 or 27.0.1.1, or any later release. The fix is a four-line change to OpenJDK (JDK-8393212), and vendors ship it under their own build numbers:
| Vendor | Fixed builds |
|---|---|
| Oracle JDK | 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1 |
| Eclipse Temurin | 17.0.21.1+2, 21.0.13.1+2, 25.0.5.1+2, 27.0.1.1+4 |
| Amazon Corretto | 17.0.21.1.2, 21.0.13.1.2, 25.0.5.1.2, 27.0.1.1.4 |
| Azul Zulu | 17.70.201, 21.54.201, 25.38.201, 27.30.201 (free builds) |
| Red Hat | Fixed packages for the Java versions each RHEL release ships (RHSA-2026:100563 and later); Red Hat build of OpenJDK 17.0.21.1, 21.0.13.1, 25.0.5.1 |
For other builds, such as Microsoft Build of OpenJDK, BellSoft Liberica and SAP SapMachine, look for CVE-2026-114938 in the vendor's release notes. Software that bundles its own Java needs its vendor's update. CISA advises patching internet-facing systems first and watching for unexpected JVM crashes.
If you cannot update yet, start Java with one of the two options Calif and the oss-security mailing list give:
-XX:-UseBASE64Intrinsicsturns off the hardware-accelerated Base64 code. Microsoft notes that Base64 decoding may be slower.-XX:UseAVX=2turns off AVX-512. Use it only on x86-64: Java on ARM does not recognize it and will not start.
Java also reads options from the JAVA_TOOL_OPTIONS environment variable. To confirm a running Java has the first option, run jcmd <pid> VM.flags -all | grep UseBASE64Intrinsics where jcmd is available: it should show false. Then update as soon as you can.
Is Java4Shell being exploited?
Attempts have been reported, but as of November 21, 2026 none of the reports we found confirms a successful attack, and the CVE is not in CISA's KEV catalog (version 2026.11.20). By November 21 (20:18 UTC), GreyNoise had logged 261 Java4Shell-shaped requests from seven IP addresses, with no successful intrusions on its sensors; SecurityWeek reported that watchTowr, Cloudflare and Previdian also saw probing or attempts. CISA's assessment on the CVE record, made on November 17, says exploitation "none".
Calif published a safe test program that shows the memory corruption and says it will publish its product exploits later. VulnCheck says a full remote-code-execution exploit has been available to its Initial Access Intelligence customers since November 17, and SecurityWeek reports that Immersive's Kevin Breen has built a working exploit. By November 19, GitHub repositories had appeared that describe themselves as remote-code-execution exploits. We have not tested them; treat exploit code from unknown authors as untrusted.
FIRST's EPSS model put the chance of exploitation activity in the next 30 days at 0.456 (about 46%) on November 21, up from 0.009 the day before, higher than 98.8% of the vulnerabilities it scores.
Who found it?
Calif says its researcher Quang Luong found the bug using OpenAI's Codex; the New York Times reports that Calif used OpenAI's GPT-6 Cyber model. Oracle credits Quang Luong of Calif.io and Albert Yang. Calif's timeline says the crash was first reported publicly in June and that Calif found the bug independently on October 24, reporting it to Oracle that day. In September, Albert Yang had proposed a fix in an OpenJDK pull request that was closed unmerged.
Oracle shipped the fix in its scheduled November Critical Security Patch Update; it was not an emergency release.
If these facts change, we will add a dated note here. The CVE-2026-114938 page shows the KEV status, EPSS score, whether a fix is available and whether exploit code is public, as our sources report them.
Sources
Read on November 21, 2026, except where a date is given:
- Oracle, Critical Security Patch Update Advisory, November 2026
- OpenJDK Vulnerability Advisory, 2026/11/17, the fix commit and pull request #32781 (read November 18)
- The CVE record and NVD
- Calif, Java4Shell
- Red Hat, CVE-2026-114938
- CISA alert, November 17, 2026 (read November 18) and CISA's KEV catalog
- GitHub advisory GHSA-6g7j-4xh3-xq2q
- FIRST EPSS
- oss-security, November 17, 2026
- Amazon Corretto 27 changelog (read November 18), Azul Zulu release notes and Eclipse Temurin releases
- OpenJ9 issue #25216
- Microsoft, Azure Functions Java worker issue #909
- GreyNoise, Java4Shell activity
- Elastic, Java4Shell security announcement, November 18, 2026
- VulnCheck, November 17, 2026
- SecurityWeek, November 19, 2026
- The New York Times, November 17, 2026
Protect your infrastructure before the breach
Map your attack surface, automate compliance, and detect insider threats in real time.
Book a demo →