The short version
- Status on 17 November 2026: reported, not confirmed. Oracle published its November Critical Patch Update the same day, with 11 Java SE fixes, but hadn't said whether CVE-2026-114938 is one of them.
- Early reports: pre-auth RCE through memory corruption, OpenJDK 17.0.6 and later (that is, since January 2023), CVSS 4.0 9.2. Some reports trace it to the x86 AVX-512 path of the
java.util.Base64decoder. - Running Java isn't the same as being exploitable: attacker-controlled input has to reach the vulnerable code. No exploitation or public proof-of-concept had been reported.
- Do now: inventory every JDK (base images, bundled runtimes, appliances), rank internet-facing services first, and watch your JDK vendor's advisory for the fixed build number.
- Two names, one CVE: "Java2Shell" and "Java4Shell" both refer to CVE-2026-114938. It is unrelated to Log4Shell, which was a library bug.
1. What's known, and what's only reported
Java2Shell became a public story on 16 and 17 November 2026, and the sources disagree. Here's how the claims sort as of 17 November 2026:
| Claim | Status |
|---|---|
| A CVE, CVE-2026-114938, exists for an unauthenticated RCE in Java | Widely reported; watchTowr said it was aware of the reports but that the technical details were unconfirmed |
| Affects OpenJDK 17.0.6 and later | Early reports; Oracle hasn't named the component |
| CVSS 4.0 score of 9.2 | Reported. Oracle scores in CVSS 3.1, and its highest Java SE score in the November update is 8.1; whether those describe the same bug is unclear |
| Root cause: memory corruption | Reported, attributed to an advisory from South Korea's National Cyber Security Center |
Specifically an integer underflow in the AVX-512 java.util.Base64 decoder intrinsic on x86 | Attributed to Oligo in secondary coverage; not confirmed by Oracle or OpenJDK |
| Fixed in Oracle's November 2026 Critical Patch Update | Unconfirmed. The update has 11 Java SE fixes; the CVE list per fix wasn't public |
| Exploited in the wild | No reports; no public proof-of-concept; not in CISA KEV |
The researchers credited with the find, at Calif, said they would publish technical details after Oracle ships a patch. That's the moment to watch: for React2Shell a working exploit was public about a day after the details were.
2. Java2Shell, Java4Shell, and Log4Shell
Two nicknames are circulating for the same CVE. "Java4Shell" follows the 2021–2022 pattern (Log4Shell, Spring4Shell, Text4Shell); "Java2Shell" follows the 2025–2026 one (React2Shell, PHP2Shell, Node2Shell). Search for both when you read advisories.
It isn't Log4Shell again. Log4Shell was a bug in the Log4j library that you fixed by upgrading a dependency. CVE-2026-114938 is reported to be in the JDK, so the fix is a new runtime: a new base image, a new JDK package, or a vendor's appliance update. Nothing in your pom.xml will change.
3. Who's exposed
Running Java doesn't make a service exploitable by itself. If the early reports hold, exposure needs three things:
- An affected build: OpenJDK 17.0.6 or later. That covers almost every Java 17, 21 and 25 runtime built since January 2023, from any vendor that builds from OpenJDK (Eclipse Temurin, Amazon Corretto, Azul Zulu, Red Hat, Microsoft).
- The vulnerable code path: if the AVX-512 detail is right, x86 hardware with AVX-512 enabled. Many current cloud x86 instances have it; Graviton and other Arm instances don't.
- Attacker-controlled input reaching it: if it's the Base64 decoder, think of anything that decodes Base64 from a request. That includes JWTs, Basic auth headers, data URIs, SAML messages and JSON fields carrying binary data, which is most web applications.
Java also hides in places that aren't Java projects: Elasticsearch and OpenSearch, Kafka, Jenkins, Keycloak, Solr, many vendor appliances, and anything with a jre/ folder in its install directory.
4. What to do before a fixed build ships
- Inventory every JDK. Base images (
eclipse-temurin,amazoncorretto,openjdk,tomcat), JDK packages installed in Dockerfiles, Lambda Java runtimes, bundled runtimes in third-party software. Record the exact version:java -versioninside the running container, not the Dockerfile. - Rank by exposure. Internet-facing services that parse request data first; internal batch jobs last.
- Apply Oracle's November 2026 update anyway. It fixes 11 other Java SE issues, and it may include this one.
- Subscribe to your vendor's advisory. Temurin, Corretto and Zulu publish fixed builds on their own schedules. The fixed version is what you'll bump to.
- Be careful with unofficial workarounds. Capping the JVM at AVX2 (
-XX:UseAVX=2) and swapping the JDK Base64 decoder for a pure-Java one have been suggested and used defensively by some projects. Neither is confirmed as a mitigation, and both change performance. Test before relying on them. - Plan for the window. A JDK upgrade across a fleet takes days to weeks. Decide now how you'll know if someone gets in before you finish.
5. How Compass helps
Compass, our open-source (Apache 2.0) operations agent for AWS, helps here in a limited way today, and it's worth saying exactly where.
The inventory question
Compass's map records the image each ECS task definition runs and what each repository's Dockerfiles build, including openjdk, eclipse-temurin and tomcat base images, and it knows which workloads face the internet. "Which internet-facing services run a JDK 17+ image?" is a query against the map, not a week of asking teams.
The patch pull request: not yet
There's no confirmed fixed build to move to, so there's nothing for a patch pull request to bump. Java also isn't yet one of the products Compass's patch captain follows in the CVE list by default (it covers PHP, Node, Python, nginx and others today, see the patching docs). For Java2Shell, plan on your own upgrade process.
The window: honeytokens
- This is where Compass is useful now. Whatever the exploit turns out to be, code execution in a JVM leads to the same next step: read the environment and the config files for credentials. Compass plants decoy AWS keys and database logins as Terraform pull requests, for you to put next to the real ones.
- When a decoy is used, CloudTrail and EventBridge deliver the event, and Compass opens an incident naming the one task definition the decoy lives in. If your policy allows, it restarts the ECS task or deactivates the key, with a snapshot, verification, audit entry and rollback.
That's how we contained PHP2Shell in under two minutes, ten months after React2Shell took four days. We closed that talk at DevOpsDays Barcelona on 14 November 2026 with "bring on Java2Shell" as a joke. Three days later it had a CVE number. Build a honeytoken by hand · More about Compass.
6. FAQ
What is Java2Shell?
Java2Shell, also called Java4Shell, is the nickname for CVE-2026-114938, a reported remote code execution vulnerability in OpenJDK. Early reports describe it as unauthenticated memory corruption affecting OpenJDK 17.0.6 and later, with a CVSS 4.0 score of 9.2. As of 17 November 2026 the technical details had not been confirmed by Oracle or the OpenJDK project.
Is Java2Shell the same as Log4Shell?
No. Log4Shell (CVE-2021-44228) was a bug in the Log4j logging library, fixed by upgrading that library. CVE-2026-114938 is reported to be in the JDK itself, so the fix is a new Java runtime build, not a dependency bump in your application.
Which Java versions are affected by CVE-2026-114938?
Early reports say OpenJDK 17.0.6 and later, which would cover Java 17, 21 and 25 builds released since January 2023. Exposure is also reported to depend on the build, on x86 hardware with AVX-512, and on whether attacker-controlled input reaches the vulnerable code. Wait for your JDK vendor's advisory (Oracle, Eclipse Temurin, Amazon Corretto, Azul Zulu, Red Hat) for the confirmed affected and fixed builds.
Does Oracle's November 2026 Java update fix Java2Shell?
Not confirmed. Oracle's Critical Patch Update of 17 November 2026 contains 11 Java SE fixes, and reports said it hadn't yet been confirmed whether CVE-2026-114938 is among them. Other OpenJDK vendors such as Eclipse Temurin, Amazon Corretto and Azul Zulu publish on their own schedules. Apply the November update anyway, since it fixes other vulnerabilities.
Is there a workaround for Java2Shell?
No official workaround had been published as of 17 November 2026. Some projects have defensively replaced the JDK's Base64 decoder with a pure-Java one, and capping the JVM at AVX2 with -XX:UseAVX=2 has been suggested, but neither is confirmed by OpenJDK as a mitigation. Test any such change for performance before using it, and prefer limiting internet exposure of Java services until a fixed build ships.
Is Java2Shell being exploited?
No exploitation and no public proof-of-concept had been reported as of 17 November 2026, and it was not in CISA's Known Exploited Vulnerabilities catalogue. The researchers who reported it said they would publish details after a patch is available, which is typically when exploitation attempts start.
Written by Ilia Dubovskii, founder of Beamreach, on 17 November 2026, the day the reports appeared. Sources (watchTowr, SecurityOnline, Oligo as quoted by others, Oracle's November 2026 Critical Patch Update coverage) disagree on several details. This page reports what was public then and will be updated as Oracle and the OpenJDK vendors confirm the affected and fixed builds.
Related: The *2Shell vulnerabilities, 2021 to 2026 · PHP2Shell · Node2Shell · AWS canary credentials · One Fake Password Is Worth More Than 10,000 Alerts (talk)
Beamreach