A newly disclosed memory-corruption bug in OpenJDK, tracked as CVE-2026-114938 and nicknamed the Java4Shell flaw, could let unauthenticated attackers launch RCE attacks against widely used enterprise software, including Jenkins, Atlassian Jira, Confluence and Keycloak.
Java powers a large share of enterprise applications, web services, financial systems and business infrastructure. Because so many unrelated programs depend on the same runtime, a weakness in one of its core components can spread across a broad range of software.
How CVE-2026-114938 Works
The problem lies in Java’s hardware-accelerated Base64 decoding function. Base64 is used to handle encoded data that regularly appears in authentication processes, file uploads, and network communications.
Security firm Calif, which disclosed the issue on Tuesday, said specially crafted Base64 input can cause the decoder to overwrite parts of Java’s memory. That can lead to the exposure of sensitive information or to remote code execution.
According to the researchers, triggering the corruption takes very little effort. A string of 64 “A” characters followed by a single equals sign (=) is enough on affected versions, while older OpenJDK 17 builds require 128 “A” characters before the equals sign.
Calif classified the bug as critical, assigning it a CVSS 4.0 score of 9.2. Oracle rated it lower, at 8.1 (High), under CVSS 3.1.
Who Is Exposed to the Java4Shell Flaw
The vulnerability affects OpenJDK 17.0.6 and later, along with Java distributions built from the same code. These include Oracle Java SE, Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, and Red Hat OpenJDK.
There is a hardware condition, however. The flaw exists only in a decoding function that runs on certain x86-64 processors supporting AVX-512 instructions, specifically those with AVX-512 VBMI capabilities.
Applications Vulnerable to RCE Attacks
Calif reported achieving code execution against several popular applications running default configurations. Unauthenticated RCE was demonstrated on Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins, and PingFederate. Authenticated RCE was achieved on Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Spinnaker.
The team also exploited the bug in sample applications built on the Apache Struts and Spring Boot frameworks.
Still, no universal exploit exists. Attackers need an application-specific way to route malicious Base64 data to the vulnerable decoder before they can run code.
AI-Assisted Discovery
Calif said researcher Quang Luong found the bug with the help of OpenAI’s Codex AI coding assistant while auditing OpenJDK’s internals. Luong reportedly identified the flaw in around one hour and built an initial working exploit in roughly two more hours, with total AI usage costs below $200. The company added that AI agents also assisted in developing further application-specific exploit chains.
Calif has released a proof-of-concept showing the memory corruption but has withheld the weaponized exploits used against the enterprise applications.
Patches and Workarounds for CVE-2026-114938
Oracle issued fixes on November 17 in its November Critical Security Patch Update, which covers 655 vulnerabilities across its products, and urged customers to apply them as soon as possible.
Organizations unable to patch right away can disable the vulnerable feature by starting OpenJDK with the -XX:-UseBASE64Intrinsics option. Another option on x86-64 systems is -XX:UseAVX=2, which turns off AVX-512 instructions.
Both measures are temporary, and developers are advised to update their Java runtimes to the latest version.





































