CISA is urging organizations to patch Java4Shell immediately, warning that crafted Base64 input can crash affected Java runtimes and allow unauthenticated remote code execution. Oracle has released security fixes, while researchers have demonstrated application takeovers under specific hardware and application conditions.
The vulnerability, CVE-2026-114938, affects the Base64 decoder in the x86-64 HotSpot Java Virtual Machine, part of the Java Development Kit. Exploitation requires processors supporting AVX-512 VBMI, including Intel Xeon chips from the Ice Lake generation onward and AMD processors based on Zen 4 or newer architectures. ARM64 systems are not vulnerable, SecurityWeek reported.
In SecurityWeek’s report, Cisco Talos said it had not observed exploitation in the wild, though it expected probing to increase after disclosure. Researchers have demonstrated working attacks; successful exploitation depends on how an application exposes the affected decoder.
Enterprise applications reached through ordinary functions
Calif, whose researcher Quang Luong discovered the flaw, disclosed its findings on November 17 after news emerged ahead of the coordinated disclosure, SecurityWeek reported.
The company said its initial tests achieved remote code execution before authentication against Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate, all using default settings. It also demonstrated code execution by authenticated attackers against Apache Kafka Broker, Solr, Elasticsearch, Gerrit and Spinnaker. Calif said neither list was exhaustive.
Sample applications built with Apache Struts and Spring Boot were also compromised. Everyday functions, including authentication and file uploads, can provide a route to the decoder.
Exposure can extend to Java clients. An exploit shared with Calif by researcher samczsun showed that a Java application could be compromised when connecting to an attacker-controlled server over TLS, even if the application offered no services of its own.
These demonstrations do not amount to a universal exploit. Calif said the steps required vary with the application’s behavior and its use of Base64.
Memory corruption, leaks and application-specific exploits
The flaw lies in an optimized routine OpenJDK uses to decode Base64 on eligible processors. Invalid input can make the routine write beyond its output buffer, corrupting nearby heap data before Java rejects the input.
Calif said carefully controlled corruption could expose sensitive memory or lead to remote code execution. It calls the memory-leak capability “JavaBleed” and warned that exposed heap contents could include personally identifiable information, secret keys and tokens.
VulnCheck told SecurityWeek that crashing a vulnerable JVM was straightforward once it found a reachable entry point using the affected decoder. A short sequence of crafted requests reliably brought down the service in its tests.
Code execution is harder. VulnCheck said it had developed a full remote code execution exploit, but an attacker must manipulate the heap layout and find target-specific ways to leak addresses and bypass address space layout randomization. The firm expects attackers to find workable routes in particular applications, rather than easy, widespread exploitation.
Calif’s public proof of concept demonstrates memory corruption without executing code. The company said it was keeping product-specific weaponized exploit chains private to give users time to upgrade and planned to release them later.
Calif also credited AI with a substantial role in discovery and exploitation. It said Luong found the vulnerability in one hour and developed the first working exploit in another two hours, with a total token cost below $200.
Check the runtime, hardware and vendor patch
SecurityWeek reported Calif’s assessment that OpenJDK versions starting with 17.0.6 are vulnerable, including distributions based on it such as Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, Red Hat OpenJDK and Oracle Java SE. The outlet also reported that the OpenJDK advisory identifies Java 17, 21, 25 and 27 as affected.
Calif assigned a CVSS 4.0 score of 9.2. SecurityWeek reported that Oracle rates the flaw at 8.1 under CVSS 3.1, with remote exploitation requiring no authentication but having high attack complexity. Those scores describe severity, not observed attacks.
Oracle delivered fixes through its November 2026 Critical Security Patch Update, which SecurityWeek said addresses 11 Java SE vulnerabilities. Oracle recommends staying on actively supported versions and applying security patches without delay.
Availability across distributions requires attention. Endor Labs said users of other Java distributions would need their vendors’ patched builds; as of November 17, it had not found the fix in the public OpenJDK 17, 21 and 25 update repositories.
Administrators should follow CISA’s advice to inventory Java installations, including runtimes bundled with third-party applications, container images and appliances that may lack separate inventory entries. Apply relevant Java and application-vendor patches as they become available, prioritizing internet-facing systems.
Where immediate patching is impractical, CISA recommends restricting and validating Base64 input at application boundaries, enforcing input-size limits and limiting unauthenticated access. Defenders should also monitor for rogue runtimes, unexpected JVM crashes and suspicious Java processes.




