A memory-corruption flaw in OpenJDK, the open-source implementation of Java that underpins a large share of enterprise software, has prompted urgent patch advisories from the U.S. Cybersecurity and Infrastructure Security Agency and India’s CERT-In. Tracked as CVE-2026-114938 and nicknamed Java4Shell, the vulnerability was disclosed on November 17 by the security firm Calif, the same day Oracle and the OpenJDK project published fixes and CISA issued its alert. CERT-In released its own advisory a day later, on November 18.
What Java is and where it runs
Java is a general-purpose programming language used to build applications and software for computers, servers and other devices. It sits behind a wide range of everyday services: business software, websites, mobile apps and games. When someone logs in to a banking website, Java may be running on the backend to check login details, process transactions and communicate with databases. It also handles user accounts and data management in ways users never see. Because Java runs across different operating systems and hardware, the same program can be deployed on laptops, servers and cloud infrastructure without being rewritten for each environment.
When a Java program is written, it is converted into a format called bytecode, which is executed by the Java Virtual Machine (JVM). The JVM lets the same program run on different systems. The Java Development Kit (JDK) supplies the tools developers need to build Java applications, and OpenJDK is its open-source version. Many technology companies base their own Java distributions on OpenJDK, so a single flaw in that shared code can reach products from multiple vendors at once.
What the Java4Shell flaw does
Java4Shell is a memory-corruption flaw in the part of OpenJDK that decodes Base64 data. Base64 is a common method of turning computer data into text made of letters, numbers and symbols, and it is used in many applications and online processes, including some ways of handling login information and file uploads. The problem sits in an optimised version of the Base64 decoder built to process data faster on certain Intel and AMD processors. When that code receives specially prepared input, it can write data beyond the memory space reserved for it, damaging other information the application is holding.
The practical consequence is that a program meant to store information in one part of memory ends up overwriting what sits next to it. Calif’s disclosure says an attacker may be able to use this to read sensitive data such as passwords, access keys or session tokens. Under certain conditions the flaw could also allow remote code execution, meaning an attacker could make a computer or server run commands without having normal access to it. CISA warned that specially crafted Base64 input could crash the JVM and potentially let an attacker execute code remotely without authentication.
Which systems are actually exposed
The flaw does not affect every computer running Java. Calif and other security advisories say exploitation depends on the Java version, the application in use and the processor underneath. The issue concerns certain x86-64 systems that support AVX-512 VBMI instructions, specialised instructions used by some processors to speed up data processing. ARM64 systems are not affected by this particular issue, which narrows the pool of hardware at risk even before software versions are taken into account.
How researchers found the flaw
Calif researcher Quang Luong used OpenAI’s Codex to examine OpenJDK code, and the AI tool pointed to a possible problem in the Base64 decoder. Luong investigated the suggestion, tested the code and developed a proof of concept, a demonstration that a flaw can be triggered. Calif says it took about an hour to identify the vulnerability and another two hours to build the first working exploit. A proof of concept does not mean attackers have used the flaw in real-world incidents.
The researchers also used a technique called heap grooming. The Java heap is the memory area where an application stores information while it runs, and grooming involves influencing how that information is arranged so that a memory error reaches data an attacker wants to expose. Calif says this helped its team demonstrate how the flaw could leak memory and, in some test cases, run code.
The finding is another example of AI being used in cybersecurity research. AI tools can help researchers sift through large amounts of code and flag areas that warrant closer inspection, though the follow-up work of testing the suggestion and building a working demonstration still fell to a human researcher.
Calif tested the vulnerability against several widely used products, including Keycloak, Atlassian Jira, Confluence, Jenkins, JFrog Artifactory and PingFederate. The company says its tests achieved remote code execution without authentication in some configurations. Other products, including Apache Kafka, Apache Solr and Elasticsearch, required an attacker to be logged in for the demonstrations. Calif’s results do not mean every installation of these products is vulnerable or has been compromised.
What CISA told organisations to do
CISA’s November 17 alert urged organisations to patch the flaw and to check software inventories for Java Development Kits and Java runtimes, including versions bundled inside other applications, container images and devices. A container is a package holding an application and the components it needs to run, and Java tucked inside one may not show up as a separate software entry, which is how such installations get missed. CISA said organisations should apply vendor updates as soon as possible, giving priority to systems reachable over the internet. Where immediate patching is not possible, it recommended restricting access to affected services, validating incoming Base64 data and monitoring for unusual Java processes or unexpected crashes, while stressing that these are temporary steps rather than a substitute for installing the update.
India’s CERT-In advisory and the vendor patch map
CERT-In’s advisory on November 18 warned that a remote attacker could send a specially crafted request carrying malformed Base64 data and potentially execute code on a vulnerable system. The Indian agency advised organisations and individuals using affected Oracle Java SE and Oracle GraalVM products to apply the updates Oracle has provided.
Oracle published its November 2026 security update on November 17, listing CVE-2026-114938 among the vulnerabilities addressed, and the OpenJDK project issued its own advisory. Other vendors are handling updates for their own Java distributions, including Amazon Corretto and Red Hat builds. Organisations running Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica or SAP SapMachine should check the relevant vendor guidance rather than assume an Oracle update fixes every installation. The Canadian Centre for Cyber Security and the European Union’s CERT-EU have also told organisations to identify affected systems and install updates, with CERT-EU recommending priority for internet-facing applications.
The advisories matter to Indian organisations because the affected code is shared. A single OpenJDK flaw can surface in products from several vendors, and CERT-In’s guidance covers both organisations and individuals running affected Oracle Java SE and GraalVM versions. CERT-In has taken a similar line in recent months, issuing a separate advisory on PHP2Shell on October 22 that specifically warned micro, small and medium enterprises, a segment with limited security staff and long patch cycles.
How Java4Shell differs from PHP2Shell
Java4Shell follows a separate Calif discovery involving PHP, another programming language used to build websites and web applications. Those earlier vulnerabilities, grouped as PHP2Shell, could let attackers execute commands on affected servers. The U.S. National Security Agency issued an advisory on October 21 urging organisations to patch vulnerable systems or take them offline if updates could not be applied at once. Both cases involved AI-assisted research and pushed organisations to review and update vulnerable software. The key difference is reach: PHP is concentrated in websites and web applications, while Java sits inside a broader range of enterprise and backend systems, including identity management, banking, cloud services and other business software, so an OpenJDK flaw can affect products from several vendors that rely on the same underlying code.
Java’s earlier critical flaws
Java has weathered major security scares before. In December 2021, a flaw known as Log4Shell was found in Apache Log4j, a logging library used by many Java applications; tracked as CVE-2021-44228, it could let attackers run malicious code remotely by sending specially crafted requests to vulnerable systems. In March 2022, Spring4Shell was discovered in the Spring Framework, used to build Java applications, and could also enable remote code execution under specific conditions. Both episodes prompted security agencies to press organisations to find affected software and install patches.
What IT teams should do next
The cited advisories contain no confirmed evidence that Java4Shell has been used in real-world attacks. The technical details are public, however, researchers have demonstrated ways to trigger the flaw, and security agencies have urged organisations not to delay updates. For IT teams, the immediate task is to find every Java installation, including runtimes bundled inside other products, establish whether it is affected and install the appropriate vendor patch. Where patching has to wait, restricting access, validating Base64 input and watching for crashes or unusual Java processes remain stopgap measures until the update is applied.