Researchers at Calif disclosed Java4Shell (CVE-2026-114938) on November 17, a memory-corruption bug in OpenJDK's hardware-accelerated Base64 decoder that they used for unauthenticated remote code execution against Keycloak, Jira, Confluence, JFrog Artifactory, Jenkins and PingFederate. It affects OpenJDK 17.0.6 and later on x86-64 processors with AVX-512. Oracle patched it the same day in its November Critical Security Patch Update.
If your team runs Jira, Confluence, Jenkins, Keycloak or any other Java server software on its own machines, update the Java runtime under it this week. A flaw called Java4Shell lets an attacker with no login run code on several of those apps in their default setup, and Oracle shipped the fix on November 17 in its November Critical Security Patch Update.
What the flaw is
The bug, CVE-2026-114938, is a memory-corruption flaw in OpenJDK’s hardware-accelerated Base64 decoder, the code Java uses to unpack encoded data in logins, file uploads and network traffic. Security firm Calif disclosed it on Tuesday. According to BleepingComputer:
- Who is affected. OpenJDK 17.0.6 and later, plus Java builds made from the same code: Oracle Java SE, Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu and Red Hat OpenJDK.
- Which machines. Only x86-64 processors with AVX-512 VBMI instructions. The vulnerable code path runs only on that hardware.
- What Calif exploited. Unauthenticated remote code execution against Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate in default configurations. Authenticated code execution against Apache Kafka, Apache Solr, Elasticsearch, Gerrit and Spinnaker.
- What it takes. There is no universal exploit. An attacker has to find a way to feed crafted Base64 data to the decoder in each app. Calif published a proof of concept for the memory corruption and kept its app exploits private.
- How it was found. Calif says researcher Quang Luong found the bug in about an hour with OpenAI’s Codex, built a first working exploit in two more hours and spent less than $200 on AI usage.
Calif rates it 9.2 (critical) under CVSS 4.0. Oracle’s advisory scores it 8.1 (high) under CVSS 3.1, remotely exploitable without credentials, and lists Oracle Java SE 8u511b33-perf, 17.0.21, 21.0.13, 25.0.5 and 27.0.1, plus Oracle GraalVM, as affected versions. Oracle credits Quang Luong of Calif. Oracle’s November update fixes 655 vulnerabilities across its products.
What to do
- Update Java first. Install the November Oracle Java SE update, or your distribution’s fixed build of Temurin, Corretto, Zulu or Red Hat OpenJDK as it ships. Then restart every service so it loads the new runtime.
- If you can’t update today, turn off the vulnerable code. Calif recommends starting Java with
-XX:-UseBASE64Intrinsics, or with-XX:UseAVX=2to disable AVX-512, BleepingComputer reports. Both are temporary. - Check bundled runtimes. Many self-hosted apps ship their own Java. Updating the system Java does not fix an app that brings its own, so check each vendor’s release notes.
- Cloud versions are the vendor’s job. If you use Jira Cloud or a hosted Jenkins, the provider patches the runtime. Watch for their advisory.
AI makes a one-hour bug hunt the new baseline
This is our analysis. The name invites comparison with Log4Shell in 2021, and the reach is similar: one library sits under a long list of business apps. The exploitation path is narrower, because it needs AVX-512 hardware and a per-app delivery route. The detail that should change how small teams plan is the timeline. One researcher with an AI coding tool went from reading OpenJDK internals to a working exploit in an afternoon for under $200. Attackers have the same tools, and Google reported this week that an AI system turned fresh security patches into working exploits with no human involved.
For a remote team without a dedicated security person, the practical rule is to treat runtime updates like browser updates: apply them within days, and keep a list of every self-hosted app and the Java it runs on. Last week brought a critical Apache HTTP Server fix, and the pattern holds: the bug is old, the patch is ready, and the risk sits in the gap before you install it.
Related: Anthropic’s plan to publish severe bugs it finds after 30 days instead of 90 and Check Point’s PHP2Shell root hotfix.
