Java4Shell, CVE-2026-114938

Java4Shell, tracked as CVE-2026-114938, is a heap overflow in Java’s hardware-accelerated Base64 decoder. Researchers at Calif say it can let attackers without an account run code on Jira, Jenkins, Keycloak and other Java applications. Oracle and the OpenJDK project shipped fixes on Nov. 17, 2026, and CISA urged organizations to patch immediately. No exploitation in the wild has been reported.

Patch it, but read the conditions before you rank it next to Log4Shell. The vulnerable code runs only on x86-64 servers whose processors support AVX-512 VBMI, so a server’s exposure depends on its CPU as much as its Java version. The same container image can be vulnerable on one host and safe on the next, and that is the detail most patch plans will miss.

Which Java versions and servers are affected?

A server is exposed only when three conditions hold at once.

1. A vulnerable Java build. Java 17.0.6 and later on the 17 line, plus Java 21, 25 and 27 before the November fixes. Standard Java 8 and 11 don’t contain the flawed code, but the Java SE 8 Performance Pack build 8u511b33-perf does, as do recent GraalVM builds.

2. The right CPU. The vulnerable path runs only on x86-64 chips with AVX-512 VBMI: Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids and Granite Rapids, and AMD Zen 4 or newer. ARM64 servers, including AWS Graviton, don’t run it. On Linux, this command should print all four flags if the chip qualifies:

grep -m1 '^flags' /proc/cpuinfo | tr ' ' 'n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'

3. A reachable decoder. An attacker has to get malformed Base64 into Java’s standard decoder. Many servers decode Base64 before checking who you are, in HTTP Basic login headers, SAML artifacts and CSRF tokens, which is what makes the attack possible without a login.

Calif reports pre-authentication code execution under default settings on Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate, and with a login on Apache Kafka, Apache Solr, Elasticsearch, Gerrit and Spinnaker. It says the list is incomplete.

Java build Fixed release
Oracle Java SE 17 (17.0.6 and later) 17.0.21.1
Oracle Java SE 21 21.0.13.1
Oracle Java SE 25 25.0.5.1
Oracle Java SE 27 27.0.1.1
Amazon Corretto 17, 21, 25, 27 17.0.21.1.2, 21.0.13.1.2, 25.0.5.1.2, 27.0.1.1.4
Azul Zulu 17, 21, 25, 27 17.70.201, 21.54.201, 25.38.201, 27.30.201
Red Hat build of OpenJDK and RHEL 8, 9, 10 Not yet released as of Nov. 18

Products that bundle their own Java runtime, from CI servers to appliances, need an update from their vendor, not just a new system JDK.

How Java4Shell works

When a method runs often enough, Java’s JIT compiler swaps the standard Base64 decoding loop for a hand-tuned assembly routine, an “intrinsic,” that uses AVX-512 VBMI instructions. That routine miscalculates the output size when the input ends with an unexpected = padding character. Wiz gives the example of 64 A characters followed by a single =.

The result is a write of 62 or 63 bytes past the end of the output array on the Java heap. Java does reject the malformed string, but only after the write has happened. That ordering is the whole bug: input validation in your application can’t help, because the damage is done inside the decoder.

The overflowed bytes are fixed by the decoder, which is why early readings predicted a crash, not code execution. Calif’s researchers got around that by arranging heap memory so the overflow enlarges a neighboring array. They then used that array to write chosen bytes and a second array to read memory back, a step they call JavaBleed, and worked from there to code execution.

There is no universal exploit. Each application needs its own memory layout and timing, and some attempts are sensitive to network delay. A failed attempt usually crashes the JVM, and an attacker can often try again.

One uncomfortable footnote: a pull request to OpenJDK on Sept. 9 described an overflow in the same decoder and included a triggering input. It was closed unmerged in early November after eight weeks of inactivity.

Is Java4Shell being exploited, and why do the scores disagree?

As of Nov. 18, 2026, no exploitation has been reported. The flaw isn’t in CISA’s Known Exploited Vulnerabilities catalog, Canada’s Cyber Centre said it wasn’t aware of attacks, and Calif is holding back its exploit code.

Independent labs have confirmed the dangerous part, though. Upwind reproduced the memory corruption on an Intel Ice Lake server on AWS until Java crashed. Blackpoint Cyber crashed a test Spring Boot app with about 28,300 malformed login requests in roughly 20 seconds, without logging in, and used the JavaBleed technique to read a secret from memory in all 30 local runs. Neither reproduced remote code execution.

That split explains the scores:

Source Score What it emphasizes
Calif 9.2 (CVSS 4.0) Demonstrated pre-auth code execution on six apps
Red Hat 9.8 (preliminary) No login required
Amazon 8.8 JVM crash for unauthenticated users; code execution for low-privileged users in some environments
Oracle / OpenJDK 8.1 (CVSS 3.1) No login required, but hard to exploit

The practical read: a pre-auth crash of any affected Java service is easy today. Reliable code execution takes per-target work, but the researchers have shown it can be done, and the triggering input has been public since September. Treat it as urgent.

How to fix and mitigate Java4Shell

  1. Find every affected runtime. Inventory Java 17 and later, including runtimes bundled inside products. Then check the CPU on each host with the command above. Mixed fleets and autoscaling groups can change exposure from one instance to the next.
  2. Patch. Install the fixed build for your distribution from the table above, then restart Java services and rebuild and redeploy container images so the running process actually uses it. Red Hat users need the workaround until Red Hat ships fixes.
  3. If you can’t patch yet, turn off the vulnerable path. Start the JVM with -XX:-UseBASE64Intrinsics, which removes only the faster Base64 code. -XX:UseAVX=2 disables AVX-512 entirely, but use it only on x86-64, because Java on ARM64 won’t start with it. Upwind warns that -XX:DisableIntrinsic=_base64_decodeBlock on its own stops Java from starting. Remove the flag once the fixed build is in.
  4. Reduce exposure at the edge. CISA recommends validating Base64 input at application boundaries, enforcing input-size limits and limiting unauthenticated access. A reverse proxy that buffers complete requests, as nginx does by default, blocks some attacks but not all.
  5. Hunt for attempts. Look for Base64 tokens of 32 characters or more that end in = and whose length isn’t a multiple of four, especially in Authorization headers; bursts of IllegalArgumentException from java.util.Base64$Decoder; and unexplained hs_err_pid crash logs or restart storms. None of these proves a compromise, and their absence doesn’t prove you’re safe.

Frequently asked questions

What is Java4Shell (CVE-2026-114938)?

Java4Shell is a heap buffer overflow in the AVX-512 Base64 decoder of OpenJDK’s HotSpot JVM on x86-64. Malformed Base64 ending in an extra = makes the decoder write 62 or 63 bytes past its output buffer. Calif, which disclosed it on Nov. 17, 2026, demonstrated unauthenticated code execution against six widely used Java applications.

Which Java versions are affected?

Java 17.0.6 and later on the 17 line, and Java 21, 25 and 27 before the November 2026 fixes. Fixed Oracle builds are 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. The server must also have an x86-64 CPU with AVX-512 VBMI, such as Intel Xeon Ice Lake or later, or AMD Zen 4 or later.

Are Java 8 and Java 11 affected?

Standard Java 8 and 11 builds don’t contain the vulnerable code. The exception is the Java SE 8 Performance Pack build 8u511b33-perf, which is affected; Oracle’s latest Performance Pack build is 8u513.

Is CVE-2026-114938 being exploited?

No exploitation in the wild had been reported as of Nov. 18, 2026, and the flaw wasn’t in CISA’s Known Exploited Vulnerabilities catalog. Independent labs have reproduced crashes and memory reads but not remote code execution.

Is Java4Shell as bad as Log4Shell?

It’s narrower. Log4Shell could be triggered by almost any logged string on any hardware with a simple, reliable exploit. Java4Shell needs specific CPUs, a reachable Base64 decoder and target-specific exploitation work. A pre-auth crash is easy, though, so treat it as urgent.

How can I mitigate Java4Shell without patching?

Start the JVM with -XX:-UseBASE64Intrinsics to disable the vulnerable decoder path, or -XX:UseAVX=2 on x86-64 only. Add input-size limits and Base64 validation at the edge, and remove the JVM flag once you’ve installed a fixed build.

Sources

Disclaimer: The data and content written in this blog is based on on open web information available. Analysts and researchers are advised to confirm once again from original sources.

The post Java4Shell (CVE-2026-114938): Who’s Exposed and How to Fix It appeared first on Cyble.