Java4Shell is a newly disclosed critical vulnerability in OpenJDK, the most widely used Java runtime. The flaw, CVE-2026-114938, sits in the Base64 decoder and can let attackers without credentials overwrite and read memory and, in the worst case, execute their own code. Servers with newer x86-64 processors are affected. Oracle and the OpenJDK distributors released updates on 17 November 2026. The bug was found by researchers at Calif, who say AI did much of the work.
What is behind Java4Shell
Base64 is a common way of representing binary data as text. It is used in authentication protocols, file uploads and many network messages. To speed up decoding, OpenJDK uses special AVX-512 VBMI instructions on supporting processors. According to Calif’s technical write-up, the bug lives in exactly this hand-written assembly code, in the part that handles the final characters of an input.
Calif describes the trigger as surprisingly simple: 64 “A” characters followed by a single equals sign are enough; in older OpenJDK 17 builds it takes 128. For those 65 characters, OpenJDK allocates a 50-byte result array. After decoding the first 64 characters, the routine mishandles the stray “=” and calculates a negative output length. That turns into a full 64-byte write that runs 62 bytes past the end of the array. Java eventually rejects the input, but by then neighbouring objects in memory may already be corrupted.
This is notable because memory safety is one of Java’s core promises. The language checks memory accesses on behalf of developers. Parts of the runtime itself, however, are written in C, C++ and assembly, where such bugs still occur.
Which systems are affected
Java4Shell affects OpenJDK from version 17.0.6 onwards and every distribution built on it. Calif lists Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine, Red Hat OpenJDK, Oracle Java SE and Oracle GraalVM for JDK, among others. According to Red Hat’s security bulletin, OpenJDK 8 and 11 do not contain the vulnerable code path.
The second requirement is hardware: the faulty code only runs on x86-64 processors with AVX-512 VBMI support. According to Calif, that includes Intel Xeon from Ice Lake onwards (including Sapphire, Emerald and Granite Rapids) and AMD EPYC and Ryzen from Zen 4. ARM64 systems are not affected. Because the JVM picks the accelerated path automatically whenever the CPU supports it, matching systems are vulnerable out of the box, Red Hat says. In cloud environments with current instance types, this condition is likely to be met frequently.
Why web applications are particularly exposed
The flaw is in the standard library, not in a single application. What matters is whether an application decodes Base64 data controlled by an attacker. Red Hat names typical web examples: HTTP Basic credentials, SAML artifacts, CSRF tokens and MIME encoded words. All of them are routinely decoded before authentication completes.
Calif says it demonstrated unauthenticated remote code execution against Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate in their default configurations. With authentication, it succeeded against Apache Kafka, Apache Solr, Elasticsearch, Gerrit and Spinnaker. Sample applications built on Apache Struts and Spring Boot could also be attacked. Calif stresses that the list is not exhaustive. A client-side attack is possible as well: a Java program that opens a TLS connection to a malicious server can be compromised even if it exposes no service of its own.
The analysis by Datadog Security Labs adds an important caveat: crashing the application is more readily achievable than reliable code execution. For actual code execution, attackers would need to tailor their exploit to the runtime and the application. Calif likewise says there is no universal exploit, and it is holding back its product-specific exploit chains until operators have had time to update. The sources we reviewed do not mention active attacks so far.
Updates and interim mitigations
Calif rates the flaw 9.2 out of 10 under CVSS 4.0, which counts as critical. Oracle rates it 8.1, or high, under CVSS 3.1. Oracle shipped the fix as part of its November patch update. According to Datadog, citing Oracle, these are the first fixed releases:
- Java 17: 17.0.21.1 (affected from 17.0.6)
- Java 21: 21.0.13.1
- Java 25: 25.0.5.1
- Java 27: 27.0.1.1
Red Hat says it is expediting fixes for RHEL 8, 9 and 10 and its own OpenJDK builds. Importantly, runtimes bundled with applications and container images must be updated too. After updating a base image, application images need to be rebuilt and Java processes restarted.
Anyone who cannot update right away has two interim options. The JVM flag -XX:-UseBASE64Intrinsics disables only hardware-accelerated Base64 processing. Red Hat prefers this option because other AVX-512 optimizations stay in place. Alternatively, -XX:UseAVX=2 turns off AVX-512 entirely, which can slow down other workloads. That flag must only be set on x86-64 systems; a JVM on ARM64 does not recognize it and will fail to start. On Linux, /proc/cpuinfo shows whether a CPU supports the required instructions: Calif names the flags avx512f, avx512bw, avx512vbmi and bmi2, all of which must be present.
Found in an hour with AI
According to Calif, researcher Quang Luong found the flaw in about an hour and built the first working exploit in two more hours. The AI token cost was under 200 dollars. Media reports say he used OpenAI’s Codex. Many of the later exploit chains were then built almost autonomously by AI agents. It is Calif’s second major AI-assisted find after PHP2Shell.
The case fits a trend we have described several times recently: AI drastically shortens the time from code analysis to finding to exploit, as our report on the AI tool exploit_agent also shows. Calif draws its own lesson and calls for “vibe-uncoding”: code nobody needs, such as an assembly optimization for decoding a few hundred bytes, should be removed to shrink the attack surface.
Our take
Java4Shell is not a Log4Shell scenario with a universal one-line exploit, but it is serious enough to act on immediately. Java runs in banks, public agencies and countless web applications, and the affected hardware is widespread in data centers. Operators should check which Java versions run on which processors, including in containers and bundled applications, and update quickly or set the flag. Anyone running Java-based identity or developer tools such as Keycloak or Jenkins exposed to the internet should treat this as top priority. Only days ago, a flaw in Apache Shiro put another widely used Java security framework in the headlines.




