A newly disclosed vulnerability tracked as CVE-2026-114938 and dubbed Java4Shell exposes certain Java environments to memory corruption, sensitive data disclosure, service crashes, and potentially remote code execution. The flaw resides in OpenJDK’s hardware-accelerated Base64 decoder and can be triggered remotely when attacker-controlled Base64 data reaches the vulnerable Java runtime.
Calif researchers demonstrated pre-authentication RCE against widely deployed enterprise applications running default configurations, including Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins, and PingFederate. Authenticated exploitation was also demonstrated against Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Spinnaker, while sample Spring Boot and Apache Struts applications were successfully exploited as well.
The vulnerability is highly dependent on the underlying hardware. The vulnerable accelerated decoder runs on x86-64 processors supporting AVX-512 VBMI, including Intel Xeon Ice Lake and newer server platforms and AMD Zen 4 or later systems. ARM64 systems are not known to be affected by this specific flaw.
Calif assigned Java4Shell a CVSS 4.0 score of 9.2, Critical, while Oracle rates it 8.1, High under CVSS 3.1. Oracle released security updates on November 17, 2026, while several downstream Java vendors have been issuing their own patched builds on separate schedules.
Organizations can leverage SOC Prime’s AI-Native Detection Intelligence Platform to accelerate threat hunting and detection engineering around emerging vulnerabilities and adapt relevant detection logic across SIEM, EDR, XDR, and Data Lake environments.
Security teams can also use Prime Architect to transform emerging vulnerability intelligence into hunting queries, refine detection logic, and adapt behavioral detections to the telemetry available within their infrastructure.
CVE-2026-114938 analysis
CVE-2026-114938 affects OpenJDK 17 beginning with version 17.0.6 and subsequent Java 17 builds, as well as affected Java 21, 25, and 27 releases derived from the vulnerable OpenJDK code. Multiple downstream distributions inherit the issue, including Oracle Java SE, Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine, and Red Hat OpenJDK. Standard Java 8 and Java 11 do not contain the vulnerable code path, although Oracle’s Java SE 8 Performance Pack build 8u511b33-perf is specifically listed as affected.
The vulnerability originates in the optimized implementation of java.util.Base64.Decoder. On supported x86-64 CPUs, the JVM can automatically select AVX-512 VBMI accelerated instructions on supported x86-64 processors to improve Base64 decoding performance. Specially malformed Base64 input causes the decoder to miscalculate the amount of data that should be written to the destination buffer, resulting in a heap-based out-of-bounds write.
According to Calif’s technical analysis, one malformed input pattern causes the accelerated routine to write approximately 62 bytes beyond the allocated output buffer. Java ultimately rejects the malformed Base64 value, but the memory corruption occurs before the validation failure stops processing.
This behavior is particularly dangerous because Base64 decoding occurs throughout enterprise Java applications. HTTP Basic authentication headers, SAML artifacts, CSRF tokens, encoded file uploads, MIME content, and other application data may all pass through the standard Java decoder. Importantly, many of these operations can take place before a user is authenticated, creating pathways toward pre-authentication exploitation.
Researchers initially faced an important limitation: the bytes written beyond the buffer are largely fixed rather than freely attacker-controlled. Calif developed an application-dependent exploitation technique that manipulates the JVM heap layout so the overflow corrupts neighboring arrays. By enlarging one array and using another to read memory, researchers created a memory disclosure primitive called JavaBleed, which could then be extended toward controlled memory modification and code execution.
There is therefore no single universal Java4Shell exploit. Exploitation varies according to the application, its Base64 processing path, JVM heap layout, network conditions, and timing. Researchers noted that remote heap grooming can require repeated attempts and careful control over allocation behavior. Failed exploitation can instead terminate the JVM and cause application outages.
The CVE-2026-114938 PoC publicly released by the researchers demonstrates the underlying memory corruption, while the weaponized application-specific exploit chains used to achieve RCE against major enterprise products have been withheld. Other security teams have independently reproduced memory corruption, JVM crashes, and memory disclosure, but some have not reproduced the full remote code execution chain.
The hardware requirement significantly changes exposure assessment. A Java application running the same container image may be vulnerable on one server and unaffected on another depending on the CPU presented to the JVM. Known susceptible processor families include Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids, and Granite Rapids, along with AMD Zen 4 and newer CPUs. ARM64 environments are not known to execute the vulnerable AVX-512 code path.
Administrators can verify CPU capability on Linux by checking whether the processor exposes AVX-512 VBMI support. They should also run java -version inside the actual application environment, including containers and software packages that bundle private Java runtimes, rather than relying only on the system-wide Java installation.
For CVE-2026-114938 detection, defenders should monitor application, JVM, web, authentication, and endpoint telemetry together. Potential indicators include unusual malformed Base64 input, repeated Base64 decoding exceptions, JVM crashes, unexpected Java process restarts, and suspicious child processes launched by Java services after inbound requests.
Security teams should pay particular attention to Base64 tokens whose lengths or padding differ from expected application behavior, especially when they appear repeatedly in authentication endpoints or other unauthenticated request paths. Bursts of IllegalArgumentException errors involving java.util.Base64$Decoder may also indicate probing, although such events are not sufficient by themselves to confirm exploitation.
To Detect CVE-2026-114938 exploitation beyond the initial memory-corruption attempt, defenders should correlate suspicious inbound requests with subsequent Java process behavior. Unexpected shell execution, new outbound network connections, file downloads, persistence creation, credential access, or unusual commands launched by a Java service should be treated as high-priority evidence of potential compromise.
Traditional CVE-2026-114938 IOCs such as fixed attacker IP addresses, domains, or malware hashes are currently of limited value because the flaw is a vulnerability rather than a campaign tied to known malicious infrastructure. Behavioral detection and exposure assessment therefore provide stronger coverage than static blocklists.
As of November 18, 2026, the Canadian Centre for Cyber Security and the reviewed security reports state that no active exploitation has been confirmed. The vulnerability was also not listed in CISA’s Known Exploited Vulnerabilities catalog when The Hacker News checked on November 18. However, public knowledge of the memory-corruption primitive significantly increases the urgency of patching internet-facing Java applications.
Security professionals seeking further details for CVE-2026-114938 should track advisories from their specific Java vendor because downstream OpenJDK distributions, bundled application runtimes, containers, and commercial products may receive patches on different schedules.
CVE-2026-114938 Mitigation
The primary remediation is to update affected Java runtimes to a fixed release as soon as possible. Oracle issued its November 2026 security updates on November 17 and lists the following fixed Java SE builds:
| Java release | Affected Oracle build | Fixed build |
| Java SE 17 | 17.0.6 through | 17.0.21.1 |
| Java SE 21 | 17.0.21.0.3 | 21.0.13.1 |
| Java SE 25 | Up to 21.0.13.0.3 | 25.0.5.1 |
| Java SE 27 | Up to 25.0.5.0.3 | 27.0.1.1 |
| Java SE 8 Performance | Up to 27.0.1.0.3 | 8u513 Performance |
| Pack | 8u511b33-perf | Pack |
GraalVM users should likewise move to the corresponding November security releases. Downstream OpenJDK distributions use their own versioning and patch schedules, so organizations should verify fixes with the vendor supplying the JVM rather than assuming Oracle build numbers apply universally.
At the time of writing, Amazon had released corrected Corretto builds and Azul had published updated Zulu releases, while Red Hat was still preparing fixes for affected RHEL and OpenJDK packages. That status can change quickly, making vendor advisories the authoritative source for patch availability.
Organizations should inventory Java not only on traditional servers but also inside containers, vendor applications, appliances, CI/CD systems, identity infrastructure, and enterprise development platforms. Products that package their own JVM may need an application vendor update instead of replacing the runtime manually.
After upgrading Java, restart affected services and verify that the running JVM actually uses the patched build. Containerized applications should be rebuilt and redeployed using an updated base image or vendor runtime.
When immediate updates are impossible, administrators can temporarily disable the vulnerable accelerated decoder by starting affected JVMs with:
-XX:-UseBASE64Intrinsics
This is the preferred temporary workaround because it disables the hardware-accelerated Base64 implementation while leaving other AVX-512 optimizations available.
An alternative is:
-XX:UseAVX=2
This prevents the JVM from using AVX-512 entirely and therefore removes the vulnerable code path. This option should only be applied to x86-64 environments; ARM64 Java runtimes do not recognize the flag and may fail to start.
Additional defensive measures should prioritize internet-facing Java applications, identity systems, authentication portals, CI/CD platforms, and development infrastructure because these environments commonly process attacker-controlled Base64 before authentication.
Organizations should also monitor Java services for unexpected crashes or restarts during the remediation period. Because failed exploitation can terminate the JVM, repeated application failures combined with malformed inbound requests may warrant immediate incident-response investigation rather than being treated as ordinary availability issues.
FAQ
What is CVE-2026-114938 and how does it work?
CVE-2026-114938, also known as Java4Shell, is a memory-corruption vulnerability in OpenJDK’s hardware-accelerated Base64 decoder. On x86-64 processors supporting AVX-512 VBMI, malformed Base64 input can cause the decoder to write beyond an allocated heap buffer. Researchers demonstrated that careful manipulation of neighboring heap objects can transform this corruption into memory disclosure and, for specific applications, remote code execution.
When was CVE-2026-114938 first discovered?
The precise date of the initial private discovery has not been publicly specified. Calif researcher Quang Luong discovered the flaw while auditing OpenJDK HotSpot with AI-assisted tooling and reported that identifying the bug took roughly one hour, followed by approximately two additional hours to develop an initial exploit. The vulnerability was publicly disclosed and patched on November 17, 2026.
What is the impact of CVE-2026-114938 on systems?
Successful exploitation can corrupt JVM memory, crash Java applications, expose sensitive information from process memory, or enable arbitrary code execution in the security context of the affected Java process. Calif demonstrated unauthenticated RCE against Keycloak, Jira, Confluence, Artifactory, Jenkins, and PingFederate, with additional authenticated attack paths against several other Java products.
Can CVE-2026-114938 still affect me in 2026?
Yes. An environment remains potentially exposed if it runs a vulnerable OpenJDK-derived runtime, executes on an x86-64 CPU supporting AVX-512 VBMI, and passes attacker-controlled Base64 data into Java’s standard decoder. Internet-facing applications and authentication services warrant particular attention. Standard Java 8 and Java 11 are not affected by this code path, while the specific Oracle Java 8 Performance Pack build identified by the vendor is vulnerable.
How can I protect myself from CVE-2026-114938?
Update affected Java runtimes using the security release provided by your Java or application vendor and restart all impacted services. Inventory bundled and containerized JVMs, not just system Java installations. If patching cannot be completed immediately, temporarily disable the Base64 intrinsic with -XX:-UseBASE64Intrinsics or disable AVX-512 on x86-64 systems with -XX:UseAVX=2, then remove the workaround after deploying a permanent fix.