Java4Shell (CVE-2026-114938): High-Severity OpenJDK vulnerability and how to patch it

Yuval Saggie
Yuval Saggie
Nov 18, 2026 | 10 min read
Java4Shell (CVE-2026-114938): High-Severity OpenJDK vulnerability and how to patch it

Key takeaways

  1. Java4Shell (CVE-2026-114938) is a high-severity memory corruption vulnerability in OpenJDK's Base64 decoder that can allow unauthenticated attackers to execute code remotely or access sensitive information.
  2. Researchers demonstrated remote code execution against widely used enterprise applications, including Jenkins, Jira, Confluence, Keycloak, and JFrog Artifactory, running in default configurations.
  3. The vulnerability affects unpatched OpenJDK versions from 17.0.6 onward, including Java 21, 25, and 27, on x86-64 processors supporting AVX-512 VBMI.
  4. AI-assisted research uncovered the vulnerability in approximately one hour and produced an initial working exploit in another two hours, demonstrating how quickly previously unknown vulnerabilities can become exploitable.
  5. Echo shipped patched Amazon Corretto container images within hours of the upstream fixes becoming available. Customers can re-pull their existing tags and redeploy to receive the fix.

Java4Shell puts widely used enterprise applications at risk

On November 17, 2026, security researchers at Calif disclosed Java4Shell (CVE-2026-114938), a high-severity vulnerability in OpenJDK that could allow attackers to execute malicious code remotely without authentication.

The vulnerability affects Java's Base64 decoder, a core component applications use to process encoded data. Researchers demonstrated successful exploitation against default configurations of widely deployed enterprise software, including Jenkins, Jira, Confluence, Keycloak, and JFrog Artifactory.

What makes Java4Shell particularly concerning is where the vulnerability lives. Rather than affecting a specific application or third-party dependency, it exists in the Java runtime itself. Applications that rely on vulnerable Java distributions may therefore be exposed through functionality they already use, without requiring unusual configurations or additional software.

The discovery also highlights the growing impact of AI on vulnerability research. According to Calif, its researchers used AI assistance to identify the flaw in approximately one hour and develop an initial working exploit in another two hours, at a total AI token cost of less than $200.

The OpenJDK Vulnerability Group confirmed the flaw, and Oracle released fixed Java builds on November 17, with other OpenJDK distributions following on their own schedules. As of November 18, there are no confirmed reports of exploitation in the wild, but the technical details are public, and researchers have already demonstrated working attacks against major enterprise applications.

What is Java4Shell (CVE-2026-114938)?

Java4Shell is a memory corruption vulnerability in OpenJDK's hardware-accelerated Base64 decoder, tracked as CVE-2026-114938.

Base64 encoding is commonly used to represent binary data as text. Java applications encounter it in everyday operations, including authentication, file uploads, API requests, and network communications.

To improve decoding performance, OpenJDK uses specialized processor instructions on supported x86-64 systems. The vulnerable implementation uses AVX-512 VBMI, an instruction set available on certain modern Intel and AMD processors.

The problem occurs when the decoder processes specially crafted, malformed Base64 input. An error in the optimized decoding routine can cause data to be written beyond its allocated memory buffer, corrupting nearby memory.

Java normally provides memory safety protections that prevent applications from accessing memory incorrectly. However, this particular optimization is implemented in low-level assembly code, where those protections do not prevent the underlying memory corruption.

An attacker who can cause an application to process malicious Base64 input may be able to exploit the corruption to crash the Java process, disclose sensitive information such as credentials and encryption keys, or execute arbitrary code.

Researchers also demonstrated a memory disclosure technique they call JavaBleed, which can expose sensitive information stored in the application's memory.

Successful remote code execution is more complex than triggering the initial memory corruption. There is no universal exploit that works against every Java application, and attackers must adapt their techniques to the application's behavior and memory layout. Nevertheless, researchers have demonstrated that exploitation is possible against several widely used products.

How severe is CVE-2026-114938?

Java4Shell has received two different severity assessments:

  • CVSS 4.0: 9.2 (Critical), according to Calif's researcher assessment.
  • CVSS 3.1: 8.1 (High), according to Oracle.

The difference reflects the use of different scoring frameworks and assessments of exploit complexity. Both ratings recognize the potential for serious consequences, including remote code execution without authentication.

Which Java versions and applications are affected?

Java4Shell affects unpatched OpenJDK builds starting with version 17.0.6, including affected releases across Java 17, 21, 25, and 27.

Because multiple Java distributions share the underlying OpenJDK codebase, the vulnerability extends beyond Oracle's own runtime. Potentially affected distributions include Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, Red Hat OpenJDK, BellSoft Liberica, SAP SapMachine, and Oracle Java SE, and Oracle GraalVM.

Certain Oracle Java 8 Enterprise Performance Pack builds are also affected, although standard OpenJDK 8 and 11 releases are not included in the affected range.

Which applications are vulnerable to Java4Shell?

Calif demonstrated unauthenticated remote code execution against several enterprise applications running in their default configurations:

  • Jenkins
  • Atlassian Jira
  • Atlassian Confluence
  • Keycloak
  • JFrog Artifactory
  • PingFederate

Researchers also demonstrated authenticated remote code execution against Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Netflix Spinnaker.

These lists are not exhaustive. Calif also demonstrated remote code execution against sample Spring Boot and Apache Struts applications, so applications built using these frameworks may also be exposed when attacker-controlled input reaches the vulnerable decoder.

Importantly, the vulnerability does not require an application to expose a dedicated Base64 decoding feature. Common functionality such as login processing, token validation, and file uploads may provide an entry point.

Which processors are affected?

Java4Shell requires an x86-64 processor supporting AVX-512 VBMI instructions.

Affected processor families can include Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids, and Granite Rapids, as well as AMD Zen 4 and newer processors.

ARM64 systems and x86-64 processors without AVX-512 VBMI support are not affected by the vulnerable hardware-accelerated decoding path.

Organizations should verify both the Java runtime version and the processor capabilities of their deployments. An application running on unaffected hardware today may become exposed if it is later deployed to a different cloud instance or server.

How AI accelerated the discovery and exploitation of Java4Shell

One of the most significant aspects of Java4Shell is how quickly researchers were able to move from vulnerability discovery to a working exploit.

According to Calif, researcher Quang Luong used OpenAI's Codex to investigate security-sensitive areas of OpenJDK, including its compiler, cryptographic functionality, and widely accessible APIs.

The AI-assisted investigation identified the Base64 memory corruption vulnerability in approximately one hour. Developing the initial working exploit took another two hours, with total AI token costs below $200.

The researchers combined AI-generated analysis with human expertise, custom tooling, and exploit-development infrastructure. Subsequent exploitation work was further accelerated through AI agents.

This matters because vulnerability discovery and exploitation have historically required substantial specialized knowledge, time, and resources. AI is making parts of that process significantly faster.

Java4Shell illustrates how a previously unknown vulnerability in foundational software can move from discovery to demonstrated exploitation within hours. Once a vulnerability becomes public, other researchers and attackers can build on the same information and increasingly capable AI tools.

For security teams, this puts additional pressure on an already difficult remediation process. Discovering vulnerabilities faster does not automatically mean organizations can patch them faster. Every affected runtime, application, and container image still needs to receive the correct fix.

As AI accelerates vulnerability discovery, the time between disclosure and potential exploitation becomes an increasingly important consideration. Organizations need a reliable way to deliver patched software quickly, without depending on lengthy manual remediation processes.

How to detect and patch Java4Shell

The most effective way to remediate CVE-2026-114938 is to update affected Java runtimes to fixed versions provided by their respective vendors.

Oracle released patched Java builds on November 17, 2026, including:

Java release Fixed Oracle Java SE version
Java 17 17.0.21.1
Java 21 21.0.13.1
Java 25 25.0.5.1
Java 27 27.0.1.1
Java 8 Enterprise Performance Pack 8u513-perf

Other Java distributions follow their own release schedules and versioning conventions, so organizations should verify that their chosen vendor's build explicitly includes the fix for CVE-2026-114938.

Check your Java runtime and processor

Start by identifying the Java version used by each application:

java -version

On Linux, check whether the processor supports the vulnerable instruction set:

grep -m1 '^flags' /proc/cpuinfo | tr ' ' '\n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'

All four flags (avx512f, avx512bw, avx512vbmi, and bmi2) should appear in the output. This does not automatically mean the application is vulnerable, but it indicates that the hardware requirements for exploitation may be present.

Organizations should also inspect Java runtimes bundled inside third-party applications, container images, and appliances. Updating the host operating system alone will not necessarily update a Java runtime packaged inside an application.

Temporary mitigations for Java4Shell

If an immediate runtime update is not possible, you can disable the vulnerable hardware-accelerated Base64 decoding path using the following JVM startup flag:

-XX:-UseBASE64Intrinsics

This forces Java to use a different Base64 decoding implementation, avoiding the affected optimization.

An alternative is:

-XX:UseAVX=2

This disables AVX-512 optimizations more broadly. Apply it only on x86-64 hosts: ARM JVMs don't recognize the flag and will fail to start, so don't roll it out across mixed fleets.

Both options may affect performance, and Java processes must be restarted for the changes to take effect.

These are temporary mitigations, not permanent fixes. The underlying vulnerability remains in the runtime until a patched version is deployed.

Organizations should prioritize internet-facing Java applications, authentication services, and systems processing untrusted input, while continuing to update affected internal workloads.

How Echo protects customers against Java4Shell

Echo shipped patched Amazon Corretto container images within hours of the upstream fixes becoming available, allowing customers to receive protection against CVE-2026-114938 without manually rebuilding and maintaining patched Java runtimes.

For customers already using Echo's Amazon Corretto images, remediation is straightforward: re-pull the existing image tags and redeploy the affected workloads. Applications built on those base images should also be rebuilt and redeployed to incorporate the updated runtime.

Echo's software factory continuously monitors upstream security fixes, incorporates patches into affected artifacts, and validates the resulting builds before delivery. For Java4Shell, Echo continues to monitor additional Java distributions and incorporate fixes as they become available.

This approach reduces the operational burden of responding to vulnerabilities in foundational software. Instead of requiring engineering teams to independently track vendor releases, rebuild affected container images, and maintain their own patched versions, Echo delivers updated software through the existing container workflow.

Customers can keep their application code, development processes, and deployment tooling while receiving patched runtime components.

Java4Shell highlights the need for faster vulnerability remediation

Java4Shell demonstrates how quickly the security landscape is changing. A high-severity vulnerability in a foundational runtime was discovered with AI assistance in approximately one hour, and an initial working exploit followed just two hours later.

The vulnerability also shows how a single flaw in widely shared software can affect many otherwise unrelated enterprise applications. Even organizations with mature security programs can find themselves exposed through dependencies and runtimes that sit deep within their software supply chain.

While Java4Shell has not been confirmed as exploited in the wild as of November 18, the availability of technical details and working research exploits makes timely remediation essential.

The long-term challenge extends beyond this individual CVE. As AI makes vulnerability discovery faster and more accessible, organizations need software supply chains capable of keeping pace.

With Echo, patched software is delivered through the sources engineering teams already use, reducing the time and effort required to address newly disclosed vulnerabilities.

For Java4Shell, Echo customers using affected Amazon Corretto images can pick up the available fix by re-pulling and redeploying. For the next critical vulnerability, the goal remains the same: make remediation part of the software supply chain rather than another emergency for engineering teams.

Learn more about Echo's frontier-hardened software.

Frontier models can now find and exploit zero days in your code

Find out what to do when there's no CVE, scanner alert or fix available yet.

What are the 7 blind spots in your vulnerability scans?

Discover when "0 vulnerabilities" doesn't actually mean you're clean.

Read now →

Ready to eliminate vulnerabilities at the source?