A critical memory-corruption vulnerability, identified as CVE-2026-114938 and dubbed Java4Shell, has been discovered in the OpenJDK platform. The flaw resides within the hardware-accelerated Base64 decoding function, which is utilized by various Java distributions on x86-64 processors that support AVX-512 instructions. By sending specially crafted Base64 data, an attacker can trigger memory corruption, potentially leading to unauthorized information disclosure or remote code execution (RCE).
The vulnerability affects OpenJDK version 17.0.6 and all subsequent versions. Numerous Java distributions are impacted by this issue, including Oracle Java SE, Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, and Red Hat OpenJDK. Because these runtimes underpin a vast array of enterprise software, the scope of the potential impact is significant, affecting both web services and critical business infrastructure.
Researchers at Calif, who publicly disclosed the flaw on November 17, 2026, assigned it a critical CVSS 4.0 score of 9.2. Oracle, however, evaluated the vulnerability as high severity with a CVSS 3.1 score of 8.1. The researchers noted that the exploitability of the bug depends on the specific hardware environment, as it requires the presence of AVX-512 VBMI capabilities. The trigger mechanism involves sending a specific sequence of characters—64 "A" characters followed by an equals sign for newer versions, or 128 "A" characters for older versions—to the vulnerable decoder.
The researchers confirmed that they successfully achieved unauthenticated RCE against several prominent applications, including Jenkins, Atlassian Jira, Atlassian Confluence, Keycloak, JFrog Artifactory, and PingFederate. Additionally, they demonstrated authenticated RCE against Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Spinnaker. The flaw was also successfully tested against sample applications developed using the Apache Struts and Spring Boot frameworks.
Despite these findings, it is important to note that there is no universal exploit for Java4Shell. Achieving successful code execution requires attackers to identify an application-specific method to deliver the malicious Base64 payload to the vulnerable decoder. While the researchers have released a proof-of-concept demonstrating the underlying memory corruption, they have not provided the weaponized exploit chains used against specific enterprise software.
The discovery of the flaw was facilitated by the use of artificial intelligence. Researcher Quang Luong utilized an AI coding assistant to audit OpenJDK internals, identifying the vulnerability in approximately one hour. The development of an initial working exploit took an additional two hours, with the total cost of AI assistance reported to be under $200. The researchers further noted that AI agents were instrumental in developing the complex exploitation chains required to target specific applications.
Oracle addressed the vulnerability on November 17 as part of its scheduled Critical Security Patch Update, which included fixes for hundreds of other issues across its product portfolio. Oracle strongly advises all customers to apply these patches immediately to remediate the risk.
For organizations unable to perform an immediate update, Calif has provided temporary mitigation strategies. Administrators can disable the vulnerable hardware-accelerated functionality by launching OpenJDK with the -XX:-UseBASE64Intrinsics flag. Alternatively, on affected x86-64 systems, administrators can restrict the use of AVX-512 instructions by setting the -XX:UseAVX=2 option. These measures are intended only as stopgaps, and a full update to the latest Java runtime remains the recommended course of action.




