Oracle patched Java4Shell (CVE-2026-114938) on 17th November 2026, but Red Hat had not yet shipped fixes for its Linux or OpenJDK builds. Calif, the firm that found the flaw, says its researchers achieved pre-login code execution on Jira, Jenkins and Keycloak.
Key Facts
- Oracle’s November security update fixes the flaw among 655 patches, and Amazon Corretto and Azul Zulu have fixed builds out too.
- Red Hat had not released fixes for Red Hat Enterprise Linux 8, 9 and 10, according to its security bulletin.
- Upwind and Blackpoint Cyber reproduced crashes and a memory leak, but neither lab has reproduced Calif’s code execution.
- Admins who can’t patch yet can start Java with -XX:-UseBASE64Intrinsics, which turns off only the faster Base64 code holding the bug.
Calif’s technical write-up traces the bug to Java’s hardware-accelerated Base64 decoder, which runs only on x86-64 chips with AVX-512 VBMI. A string of 64 “A” characters plus “=” (128 on OpenJDK 17) makes it write 62 bytes past the end of its output array. Every OpenJDK build from 17.0.6 onward carries it.
Researcher Quang Luong found the flaw with OpenAI’s Codex in about an hour, Calif says, and built a first working exploit in two more. The total token bill came in under $200.
Calif’s exploits worked without a login on default setups of Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate. Apache Kafka, Solr, Elasticsearch, Gerrit and Spinnaker fell only after login, and Calif calls both lists incomplete.
🚨We urge orgs to immediately patch the recently disclosed Oracle Java vuln CVE-2026-114938. This critical memory-corruption vulnerability, known as “Java4Shell,” affects the Base64 decoder of the x86-64 HotSpot JVM, part of the Java Development Kit. 🔗https://t.co/SwrhdPFlZF https://t.co/3QGldeAerV
— CISA Cyber (@CISACyber) November 17, 2026
Red Hat users are still waiting on builds
Oracle’s November Critical Security Patch Update lists the flaw under its Java SE Compiler component. The OpenJDK advisory marks it fixed in the same week’s releases, and Amazon and Azul list fixed builds too.
| Java build | Fixed release |
|---|---|
| Oracle Java SE 17, 21, 25, 27 | 17.0.21.1, 21.0.13.1, 25.0.5.1, 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 (free builds) | 17.70.201, 21.54.201, 25.38.201, 27.30.201 |
| Red Hat Enterprise Linux 8, 9, 10 and Red Hat OpenJDK 17, 21, 25 | None yet |
Eclipse Temurin, Microsoft’s OpenJDK build and BellSoft Liberica hadn’t listed fixes yet.
Two labs crash Java, neither reaches code execution
Security company Upwind ran the trigger on an AWS c6i instance with an Intel Ice Lake chip. With an exact-size output buffer, the overflow corrupted the next object in 97% of rounds and eventually crashed Java, its lab notes show.
Blackpoint Cyber’s Adversary Pursuit Group crashed a test Spring Boot app with about 28,300 malformed login requests in roughly 20 seconds, never logging in. Its local test of JavaBleed, Calif’s memory-read step, pulled a secret stored beside the overflow in all 30 runs. It did not reproduce code execution.
Outside labs have shown a crash and a memory leak. Code execution still rests on Calif’s own testing, and the firm is keeping its product exploits private. Nobody has reported attacks, and CISA had not added the flaw to its Known Exploited Vulnerabilities catalog as of 18th November.
Severity scores reflect the same uncertainty. Oracle rates it 8.1 under CVSS 3.1 with no login needed, Calif 9.2 under CVSS 4.0 and Red Hat a preliminary 9.8. Amazon’s 8.8 describes a crash for unauthenticated users, or code execution by an authenticated low-privileged user in certain environments. Teams that rank patches by CVE severity scores will get a different answer depending on whose number they use.
The trigger sat in a public pull request
A malformed input that triggers the overflow appeared publicly on 9th September. It sat in a proposed fix submitted to OpenJDK’s code repository for the same AVX-512 Base64 decoder. Its author closed it unmerged after an automated notice flagged more than eight weeks of inactivity.
Calif said on X that details had leaked before the patch. Did anyone reviewing OpenJDK’s pull requests flag that proposal as a security bug? And is it the leak Calif cited? Neither has a public answer.
The vulnerable code runs on Intel Xeon Ice Lake through Granite Rapids chips and AMD Zen 4 or newer, never on ARM64. Calif says a Java program that connects to a malicious TLS server “can be compromised, even if it exposes no service of its own.“
Teams on Java 17.0.6 or later, including 21, 25 and 27, can run four checks today:
- Confirm whether /proc/cpuinfo lists all four flags the bug needs: avx512f, avx512bw, avx512vbmi and bmi2.
- Install the fixed build, restart Java services and rebuild container images. Products that bundle their own Java need an update from their maker.
- Until a fix lands, start Java with -XX:-UseBASE64Intrinsics. Upwind warns that -XX:DisableIntrinsic=_base64_decodeBlock on its own stops Java from starting.
- Search logs for Base64 tokens of 32-plus characters ending in “=” whose length isn’t a multiple of 4, and for bursts of java.util.Base64$Decoder errors.
Red Hat’s fixes, and the container rebuilds that follow them, are the next marker. After that, the clock belongs to Calif, which plans to publish its product exploit chains once users have had a chance to upgrade.
