Key takeaways for CVE-2026-114938
- CVE-2026-114938, also called Java4Shell, is a memory-corruption flaw in OpenJDK's hardware-accelerated Base64 decoder. According to Calif's technical write-up, it can crash the JVM or expose heap memory and, under application-specific conditions, lead to remote code execution (RCE).
- Exposure depends on the runtime build, CPU features, JVM configuration, and whether attacker-controlled Base64 data reaches the decoder. The reported vulnerable path requires x86-64 with the necessary AVX-512 features, and ARM64 systems are not affected by this path.
- Oracle rates the flaw 8.1, High, under CVSS 3.1. Bitsight CTI assigns it a DVE score of 9.99/10, Critical. CVSS rates technical severity. DVE predicts the likelihood of exploitation within 90 days based on threat intelligence.
- Calif demonstrated RCE in specific Java applications, but there is no single exploit that applies to every Java service. Its public proof of concept demonstrates memory corruption, and that alone does not establish universal remote RCE.
- Oracle and OpenJDK released fixes on November 17. Update to a vendor-confirmed fixed runtime. If you cannot patch immediately, temporarily disable the vulnerable Base64 intrinsic and restart affected JVMs.
What happened?
Calif disclosed Java4Shell on November 17. The flaw is in OpenJDK HotSpot's optimized x86-64 Base64 decoding routine. Specially crafted input can corrupt JVM heap memory, potentially causing a crash, exposing information, or enabling code execution with the Java process's privileges. Oracle lists affected builds including Java SE 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3, and 27.0.1.0.3, as well as affected GraalVM builds. The broader affected OpenJDK lines include unpatched versions of Java 17, 21, 25, and 27. The vulnerable code is present in OpenJDK 17 from version 17.0.6. Standard OpenJDK 8 and 11 builds are not affected by this specific flaw. Vendor builds and fix schedules vary, so check the advisory for the runtime you use.
Calif reported pre-authentication RCE demonstrations in default configurations of Keycloak, Jira, Confluence, JFrog Artifactory, Jenkins, and PingFederate. It also reported post-authentication demonstrations in products including Apache Kafka, Apache Solr, Elasticsearch, Gerrit, and Spinnaker.
How CVE-2026-114938 (Java4Shell) works
OpenJDK uses AVX-512 instructions to accelerate Base64 decoding on supported x86-64 processors. In the vulnerable routine, malformed padding can cause an output-length calculation to underflow. The resulting invalid write can overwrite memory beyond the decoder's output buffer. The Java code may reject the malformed input afterward, but the memory corruption may already have occurred.
The vulnerability is not reachable on every Java system. The reported path requires a vulnerable runtime, the affected intrinsic enabled, a compatible x86-64 CPU exposing the required instruction features, and attacker-controlled data reaching the decoder. Calif lists Intel Xeon Ice Lake and later server families, and AMD Zen 4 or later, among the processors that meet those requirements. ARM64 systems are not affected by this reported path.
A remote attacker may be able to supply that data through an application feature that decodes Base64, such as authentication, token handling, or file processing. Calif also reported a client-side scenario in which a Java program connecting to a malicious server could be affected. Whether exploitation is possible before authentication, and whether it can lead to RCE, depends on the application's data flow and memory layout.
Understanding severity and exposure
- Oracle rates the flaw 8.1, High, under CVSS 3.1
- NVD carries the same v3.1 score
- Calif assigns it 9.2 under CVSS 4.0.
- Bitsight DVE score of 9.99/10, Critical, up from 5.75 on November 1.
- DVE models exploitation activity based on threat intelligence and predicts the likelihood of exploitation over a 90-day period.
VulnCheck said the flaw was not known to be exploited in the wild, while reporting that it had a full RCE exploit available to its customers. Bitsight CTI also did not flag exploitation in the wild, Metasploit use, or ransomware or APT association. Calif reported application-specific demonstrations and said it was withholding product-specific exploit chains. CIRCL reported that the Canadian Centre for Cyber Security had no known active exploitation in its November 17 alert. None of these reports indicate widespread attacks in the wild.