A single short string of text can break the security of millions of Java servers. That is the warning behind Java4Shell, a critical flaw disclosed on November 17, 2026, that lets an attacker run code on a vulnerable server without logging in first. Oracle and OpenJDK released emergency fixes the same day, and the United States Cybersecurity and Infrastructure Security Agency, known as CISA, has urged organisations to patch immediately.
The Java4Shell vulnerability is tracked as CVE-2026-114938. It sits not in one application, but in the Java runtime itself — the software behind a huge share of the world’s business systems. No application has to contain a bug of its own for a server to be exposed.
What the Java4Shell vulnerability is
Java4Shell is a heap-based out-of-bounds write in the hardware-accelerated Base64 decoder of the HotSpot Java Virtual Machine, or JVM. Base64 turns binary data into plain text, and it is used almost everywhere: login headers, security tokens, cookies, file uploads and data sent between services.
The problem is in a hand-written assembly routine that speeds up Base64 decoding on modern x86-64 processors, using instructions called AVX-512 VBMI. Given specially crafted malformed Base64 input, it writes 62 bytes past the end of its output buffer, according to security firm Qualys, before Java rejects the input as invalid.
That small overflow corrupts nearby memory in the Java heap. On its own, corrupting memory usually just crashes a program. But the researchers who found the flaw showed that, in the right conditions, the corruption can be steered — turning a crash into remote code execution, where the attacker runs their own commands on the server.
The flaw was reported by security firm Calif (researchers Quang Luong and Albert Yang). Calif rates it 9.2 out of 10, or Critical; Oracle rates it 8.1 out of 10, or High. Both agree: an attacker needs no credentials, and the impact can be high.
Which systems are affected
Three conditions must all be true for a system to be at risk.
First, the Java version. The flaw affects OpenJDK builds from version 17.0.6 onward, including the Java 17, 21, 25 and 27 release lines, from any vendor — Oracle, Eclipse Temurin, Amazon Corretto, Azul Zulu, Microsoft, Red Hat, BellSoft Liberica and SAP SapMachine among them. Oracle’s Java SE 8 Enterprise Performance Pack build is also affected, as is Oracle GraalVM. Standard Java 8 and Java 11 are not affected, because the vulnerable code was introduced later.
Second, the processor. The vulnerable code path runs only on x86-64 processors with AVX-512 VBMI support: Intel Xeon chips from Ice Lake onward, and AMD processors from Zen 4 onward, including EPYC server chips and Ryzen 7000 and 9000 desktop chips. Machines running ARM64 processors, such as AWS Graviton servers and Apple silicon Macs, are not affected, and neither are older Intel chips without AVX-512.
Third, the application must pass attacker-controlled data to Java’s Base64 decoder. In practice, that is very common: any web service that accepts tokens, file uploads, encoded headers or API payloads is likely to decode Base64 at some point.

The products researchers broke into without a login
What turned Java4Shell from a worrying bug into a headline was the demonstration. Calif says it achieved remote code execution in the default configurations of six widely used Java products — without logging in at all: Keycloak, the identity and access management system; Atlassian Jira and Atlassian Confluence, the project-tracking and collaboration tools used by countless companies; JFrog Artifactory, a repository for software packages; Jenkins, the automation server at the heart of many software build pipelines; and PingFederate, an identity federation product.
The researchers also reported code execution after authentication in Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit and Netflix Spinnaker. Wiz, the cloud security company, additionally lists Apache Tomcat among the products where remote exploitation was shown to be feasible.
These are not obscure tools: Keycloak guards logins, Jenkins builds and ships software, and Jira and Confluence hold a company’s plans and documents. Taking one over without a password means stolen credentials, stolen data, or malware planted in software builds.
There is an important limit: the researchers have not published one universal exploit for every Java application. Each target needs its own steps, and Datadog Security Labs, which reproduced the flaw, stresses that crashing the server is the more readily achievable impact. The risk is real, but no single magic string takes over every Java server.
Has Java4Shell been used in real attacks?
As of November 18, 2026, no exploitation in the wild has been reported. The Canadian Centre for Cyber Security said in its November 17 alert that it knew of no active exploitation, and security firms including Qualys and Cisco Talos had also seen no attacks in the wild as of November 17. The flaw is not in CISA’s Known Exploited Vulnerabilities catalogue.
Nobody in security expects that calm to last. The technical details are now public, a proof of concept exists, and the affected products are so widely used that attackers have every reason to try. When the Log4Shell flaw appeared in 2021, attackers began scanning within hours. Java4Shell sets up the same race: defenders patching against attackers scanning.
How to check if you are affected
Work through these checks in order — they take minutes.
First, find every Java runtime you run — including ones bundled inside other products and container images. For each, note the exact version it reports and which vendor supplied it. Anything on OpenJDK 17.0.6 or later, up to the fixed builds below, is in scope.
Second, check the processor. On a Linux server, grep -m1 '^flags' /proc/cpuinfo shows the CPU features; look for avx512f, avx512bw, avx512vbmi and bmi2. All four must be present for the flaw to be reachable. Cloud users should check the instance type.
Third, check whether the vulnerable code path is switched on. The accelerated decoder is enabled by default on supported hardware, so assume it is on unless someone has deliberately switched it off. The command java -XX:+PrintFlagsFinal -version | grep -E 'UseBASE64Intrinsics|UseAVX ' shows the current settings. If all three checks point to an affected system, treat it as exposed.

How to patch Java4Shell
Updating the Java runtime is the only complete fix, and the patched builds were released on November 17, 2026. Oracle’s fixed builds are 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. The same fixes flowed into the OpenJDK builds from every vendor, including Eclipse Temurin, Amazon Corretto, Azul Zulu, Microsoft Build of OpenJDK and Red Hat OpenJDK, so install the fixed build from whichever vendor supplies your Java and then restart the affected services so the new runtime actually loads.
Start with internet-facing identity, build and collaboration tools — Keycloak, Jenkins, Jira, Confluence, Artifactory and PingFederate first. Give every remaining unpatched system an owner and a deadline, and track it until the patch is confirmed.
What to do if you cannot patch yet
Patching is not always instant: a fixed build may not have reached your Linux distribution yet, or a production system may need a maintenance window. Until then, start the JVM with the flag -XX:-UseBASE64Intrinsics. It switches off the accelerated Base64 routine, so decoding falls back to slower generic code without the bug. Calif and several security firms say the flag stopped the overflow on every affected version tested.
A second flag, -XX:UseAVX=2, also works — but only on x86-64 hosts, because ARM64 JVMs will fail to start with it. You can also set the Base64 flag as an environment variable, JAVA_TOOL_OPTIONS="-XX:-UseBASE64Intrinsics", then confirm with the PrintFlagsFinal command above that it took effect.
Other defences help while patching is under way: restrict public access to Java services that do not need it, put a web application firewall in front of the ones that do, and watch logs for unusual Base64 payloads or repeated crashes — often the first sign of probing. The flag and the firewall only buy time; the patch ends the risk.
What this means for everyone else
Most readers will never run a Java server, but Java4Shell still matters to them: the services you log into every day often run on Java behind the scenes. When a flaw like this appears, those companies must patch quickly, and users may see brief maintenance windows or forced logouts — a sign the patching is happening.
There is also a wider lesson: the flaw was in the runtime, not in any application — a reminder that supply-chain risk extends to the platform underneath the software you choose. Just days earlier, the OPTIMJACK attack turned trusted websites into malware delivery pages by poisoning one shared third-party script.
For individuals, the advice stays the same as after any major flaw: use unique passwords, turn on two-factor authentication wherever it is offered, and be wary of phishing emails that exploit security news. Our guide on how to spot a phishing email explains the warning signs.
What happens next
Researchers will keep testing which products are reachable, vendors will keep shipping fixes, and attackers will almost certainly start scanning for unpatched servers. If you run Java, patch now. If you manage a team, make Java4Shell today’s top security item, with a named owner and a deadline.
Java4Shell is serious but fixable: the patches exist, the workaround is one flag, and the checks take minutes. The danger now is leaving it unpatched.
