Skip to content
Java

Java

Heise.De • November 21, 2026

Eleven different bugs allow commands to be executed on other computers (remote code execution, RCE) running Java based on Oracle’s OpenJDK (version 17 or earlier). Oracle released updates Tuesday evening. The most serious Java vulnerability, with a CVSS 4.0 severity score of 9.2, is called Java4Shell by its discoverer, Cali.

Cali is a California-based IT security company. After the patches were released, Calif published its promised explanation of the problem and described the vulnerability. It is now clear that Java4Shell affects only certain processors: Intel Xeons from the Ice Lake, Sapphire Rapids, Emerald Rapids and Granite Rapids generations, as well as AMD EPYC and Ryzen CPUs from the Zen 4 generation and newer.

The authors tackle the thorny issue first: Java promises to be memory-safe, preventing memory-corruption errors so developers don’t have to worry them when writing Java code. That’s not exactly wrong, but it’s not really true, either. The software that Java runs on, OpenJDK, is written in several languages, including C, C++ and assembly — none of which are memory-safe.

The specific bug behind Java4Shell is in the decoder for Base64 strings. Apparently for performance reasons, OpenJDK developers decided to build this decoder in assembly so it could use features from the AVX512 instruction set extension for x86-64 processors.

This implementation contains a bug in how it handles the equals sign used to pad Base64 strings. A 65-character string is enough to trigger the problem: 64 instances of the letter A , followed by a = at the end, passed to the decoder. OpenJDK reserves 50 bytes for the decoded output, then misinterprets the trailing = after processing it and returns a negative length for the output string, circumventing memory management. As a result, the Base64-decoded output is written to the wrong location in memory. With OpenJDK 17, twice as many A characters must be entered.

As Calif explains, this isn’t a complete attack on its own, but the for real-world applications has been successful. Calif says it achieved remote code execution through these unauthorized memory accesses, without authentication, in widely used Java applications. These included the SSO tool Keycloak, Atlassian’s Jira and Confluence, JFrog Artifactory and Jenkins, all without requiring a user account. As authenticated users, the researchers were able to exploit the vulnerability in Elasticsearch, among other applications.

The nature of the problem limits the kinds of attacks possible: An attacker can’t simply send the string to a server that processes Base64 in some way. They must develop an attack targeting the specific Java application to exploit the memory management bug. Well-known, widely used Java programs exposed to the internet are therefore particularly at risk. Although Calif’s security researchers haven’t released proof-of-concept code for software such as Jira or Keycloak, attackers could develop suitable tools.

No Java4Shell attacks in the wild have been reported so far. That could change quickly. So install the latest updates immediately — even on systems with unaffected CPUs, since the restriction to the processors listed above applies only to Java4Shell. The updates also close 10 other security vulnerabilities that allow unauthorized code execution on remote systems.

Calif also briefly points to an exploit submitted to the company for Java clients. According to the report, a malicious server can compromise Java programs that connect over TLS, even if the client doesn’t offer any Java services. Whether this also affects only systems with certain CPUs, and what the remedy is, remain unclear. heise security has asked Calif for more information.

This article was originally published in German . It was translated with technical assistance and editorially reviewed before publication.