Tech

Major security flaw found in Java base code with help from artificial intelligence

A critical memory defect exposes widely used enterprise applications running on modern Intel and AMD processors, prompting urgent software patches across the technology sector.

A close up of a server in a server room
Tyler / Unsplash, Unsplash License

A critical software vulnerability in the core engine of Java could allow attackers to execute malicious code on popular corporate services without needing an account, cybersecurity researchers warned this week.

The flaw, tracked as CVE-2026-114938 and nicknamed Java4Shell, affects the open-source OpenJDK framework that underpins Java installations across the globe. Security research firm Calif announced the discovery on Tuesday, revealing that researcher Quang Luong identified the vulnerability in roughly one hour during a code audit with the assistance of OpenAI’s Codex coding tool.

According to Calif, developing an initial working exploit took another two hours, with the entire artificial intelligence assistance costing less than $200. The disclosure highlights how emerging automated tools are accelerating the pace of vulnerability research for both software defenders and potential adversaries.

How the vulnerability works

The security issue resides inside a performance-optimized Base64 decoding routine introduced in Java 17.0.6. The decoder speeds up processing on modern x86-64 server chips that support AVX-512 VBMI vector instructions, which include newer Intel Xeon models such as Ice Lake and Sapphire Rapids, as well as AMD Zen 4 and newer processors.

Because Base64 decoding commonly processes authentication tokens, login headers and web requests before a user identity is confirmed, attackers can reach the vulnerable routine without logging in. Providing a specially malformed string ending in an extra equals sign causes the decoder to write dozens of bytes beyond its allocated memory boundary.

Calif reported that researchers managed to arrange system memory to leverage the overflow into code execution against default installations of Atlassian Jira, Confluence, Jenkins, Keycloak, JFrog Artifactory and PingFederate. The firm also reported exploiting test applications built with Spring Boot and Apache Struts.

However, researchers emphasized that no universal attack exists. Exploiting the defect requires specific timing and memory configurations tailored to each target system.

Industry response and mitigations

Major enterprise software vendors moved quickly to evaluate and address the issue. Oracle rolled out fixes on Tuesday as part of its scheduled security update, scoring the vulnerability at 8.1 out of 10. Calif assigned the issue a critical severity score of 9.2, while Red Hat assigned an initial rating of 9.8.

Amazon updated its Corretto Java distribution, and Azul issued updates for its Zulu platform. Red Hat reported that it was still preparing patches for Red Hat Enterprise Linux and its OpenJDK builds.

Independent security teams have successfully reproduced memory corruption and data leaks in lab environments. Blackpoint Cyber confirmed it crashed a test application by sending thousands of malformed requests and retrieved sensitive data from memory, though it did not achieve arbitrary code execution.

Government authorities, including the Canadian Centre for Cyber Security, stated they had observed no evidence of active exploitation in the wild, and the issue does not appear on the known exploited vulnerabilities catalog maintained by the U.S. Cybersecurity and Infrastructure Security Agency.

For systems that cannot immediately apply patches, security analysts advise administrators to run their Java runtime environments with specific startup flags, such as disabling hardware-accelerated Base64 intrinsics, until official updates are deployed.