# Java4Shell (CVE-2026-114938) FAQ

**URL:** <https://faq.inetsoftware.de/t/java4shell-cve-2026-114938-faq/308>\
**Category:** General\
**Created:** [November 17, 2026, 10:51pm UTC](https://faq.inetsoftware.de/t/java4shell-cve-2026-114938-faq/308 "2026-11-17T22:51:33Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![GerryWeissbach](https://faq.inetsoftware.de/user_avatar/faq.inetsoftware.de/gerryweissbach/32/90_2.png) [@GerryWeissbach](https://faq.inetsoftware.de/u/GerryWeissbach)\
**Post date:** [November 17, 2026, 10:51pm UTC](https://faq.inetsoftware.de/t/java4shell-cve-2026-114938-faq/308/1 "2026-11-17T22:51:33Z")

</div>

As of November 17, 2026. This post will be updated as soon as JVM vendors release fixes.

## What is Java4Shell?

A critical vulnerability (CVE-2026-114938, CVSS 4.0: 9.2) in the Java Virtual Machine. In the Base64 decoder, which uses AVX-512 instructions, a crafted input string can trigger a heap overflow. A JVM crash is confirmed. The discoverer (Calif) claims code execution, but this hasn’t been independently confirmed.

## Are products of i‑net software GmbH affected?

Yes, from version 23.10 onward. These versions ship Eclipse Temurin 17 or later (23.10: Temurin 17, 24.10 and 25.10: Temurin 21, 26.4 and later: Temurin 25). The discoverer explicitly lists Eclipse Temurin as affected. The flaw is in the JVM, not in our code. In our source code, several places use `java.util.Base64` to decode data from HTTP authentication headers and login tokens. We haven’t yet tested whether this can trigger a crash.

You can also use your own Java runtime with our products. For such a JVM, the assessment of the JVM itself applies, regardless of the product version: Java 17 from 17.0.6 and Java 21, 25 and 27 before the fixed builds are affected. Several security vendors (Upwind, Endor Labs, Equixly) list Java 8 and 11 as not affected; the discoverer doesn’t mention Java 11.

## When is my system actually vulnerable?

All of the following must be true:

1. The JVM is Java 17 from 17.0.6, or Java 21, 25 or 27 before the respective fixed build.
2. The server has an x86-64 CPU with AVX-512 VBMI (Intel Ice Lake or later, AMD Zen 4 or later). ARM (for example AWS Graviton) and older x86 CPUs aren’t affected.
3. The Base64 intrinsics are enabled (the default).
4. Untrusted data reaches the Base64 decoder.

On Linux, check the CPU with:

```bash
grep -o -w -m1 'avx512vbmi' /proc/cpuinfo

```

and the JVM setting with:

```bash
java -XX:+PrintFlagsFinal -version | grep -E ' (UseAVX|UseBASE64Intrinsics) '

```

The code path is active if `avx512vbmi` appears, `UseAVX` is greater than 2 and `UseBASE64Intrinsics` is `true`.

## Is there a fix yet?

Oracle has announced builds ending in `.1` (17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1). Oracle hasn’t yet confirmed which of them close this vulnerability. No fixes are available yet for Eclipse Temurin and other distributions. The same version number without `.1` remains vulnerable.

As soon as patched Temurin builds appear, we will verify them and release updated product versions or Java updates. We will post an update here.

## What can I do until then?

- Restrict access from untrusted networks, starting with internet-facing systems.
- Check whether your servers use AVX-512 (see above). Without AVX-512 there is no exposure.
- Disable the vulnerable decoder as an interim measure (next question).
- Watch for JVM crashes: new `hs_err_pid*.log` files, restart loops, or SIGSEGV in the garbage collector.

## How do I disable the vulnerable decoder, and what does it cost in performance?

There are three JVM options. No JVM vendor has endorsed them as a fix; they bridge the time until the patch.

| Option | Effect | Performance impact |
| --- | --- | --- |
| `-XX:+UnlockDiagnosticVMOptions -XX:DisableIntrinsic=_base64_decodeBlock` | Disables only the vulnerable decoder (narrowest option) | Not measured |
| `-XX:+UnlockDiagnosticVMOptions -XX:-UseBASE64Intrinsics` | Disables Base64 acceleration | According to Cyberkendra, Base64-heavy workloads may get slower; not measured |
| `-XX:UseAVX=2` | Disables AVX-512 altogether | According to Equixly, cryptography and other vectorized code is affected too; not measured. Use on x86-64 only; a JVM without `-XX:+IgnoreUnrecognizedVMOptions` won’t start with it on ARM |

**Performance:** No measurements have been published so far, and we haven’t yet measured the impact on our products. After the change, monitor response times and CPU load.

**Recommended setting:** `-XX:+UnlockDiagnosticVMOptions -XX:-UseBASE64Intrinsics`. The order matters; `-XX:+UnlockDiagnosticVMOptions` must come first. Temurin 21 accepts and applies the option (checked on ARM, where the effect against the vulnerability can’t be measured); Java 17 hasn’t been tested yet.

**Where do I set the option?** In the web server configuration, in the field **Other VM arguments**. The arguments are split at spaces; quotation marks aren’t evaluated. Save the configuration and restart the server. The server starts its main JVM through a launcher process and reads the setting again on every restart.

The field isn’t enough in these cases:

1. **Cloud or database persistence instead of files:** The field is ignored. Add the option to the start command, in the Docker start or in the service.
2. **Start without the launcher process:** The PDFC command-line mode and the Designer don’t read the web server configuration. Pass the option directly at start.
3. **Separate JVM processes:** The scripts `startRecovery.sh`, `startHelp.sh`, `startDesigner.sh` and `compare.sh` start their own JVMs without the web server configuration. The Designer also runs on the user’s own Java runtime, for example through OpenWebStart. This can’t be controlled through the server configuration.

**Further notes:**

- `-XX:DisableIntrinsic=_base64_decodeBlock` only works together with `-XX:+UnlockDiagnosticVMOptions`. Without it, the JVM won’t start (checked with Temurin 21). Test the setting on a single server before rolling it out broadly, especially when distributing it through `JAVA_TOOL_OPTIONS`.
- The options only take effect after a server restart.
- Verify the effective setting with `-XX:+PrintFlagsFinal`.
- Remove the option once the patch is installed.

## When am I not affected by operation mode?

If your servers run on ARM or their CPU lacks AVX-512 VBMI, the vulnerable code path isn’t used.

## Is the vulnerability being exploited?

As far as currently known (November 17, 2026), no attacks have been observed. The vulnerability isn’t in the CISA KEV catalog, and there is no public exploit. This can change quickly.

## Where do I find updates?

In the news and in this FAQ post. We update both as soon as patched JVM builds are available.

## Sources

- Upwind: [https://java4shell.org/](https://java4shell.org/)
- JFrog Security Research: [Java4Shell: What We Know About the Critical OpenJDK RCE Vulnerability (CVE-2026-114938) - JFrog Security Research](https://research.jfrog.com/post/java4shell-vulnerability/)
- Endor Labs: [Java4Shell (CVE-2026-114938): Pre-Auth Heap Overflow in OpenJDK | Endor Labs](https://www.endorlabs.com/learn/java4shell-unauthenticated-rce-reported-in-openjdk-17-cve-2026-114938)
- Cyberkendra: [Java4Shell (CVE-2026-114938): Pre-Auth RCE Flaw in Java 17 Puts Servers at Risk - Cyber Kendra](https://www.cyberkendra.com/2026/11/cve-2026-114938-pre-auth-rce-flaw-in-java-17-puts-servers-at-risk.html)
