Java4Shell patches released: which Java systems need checking
Oracle and OpenJDK published fixes for CVE-2026-114938 on Nov. 17. The Base64 flaw can enable remote code execution, but exposure depends on the runtime, hardware and application.

Oracle and OpenJDK released security updates on Nov. 17, 2026, addressing Java4Shell, a vulnerability that can enable remote code execution in Java applications under specific conditions. Tracked as CVE-2026-114938, the flaw was disclosed by security company Calif and involves an optimized routine for decoding Base64 data.
The immediate concern is for organizations operating enterprise Java applications, including identity services and development infrastructure. Establishing exposure requires more than finding Java on a server: administrators need to identify the runtime build, the processor and the application’s route for processing outside input.
Oracle’s official risk matrix lists the vulnerability as remotely exploitable without authentication, with high attack complexity. Its November update fixes multiple security issues, so the broad list of products receiving patches should not be mistaken for the affected-version list for this particular flaw.
A common data conversion reaches unsafe code
Base64 represents binary data using text characters. Applications use it in network messages, file transfers and authentication-related data. It is an encoding method rather than encryption; information does not become confidential simply because it has been converted to Base64.
Calif says the vulnerability sits in machine-level code designed to accelerate decoding. A malformed ending can cause the routine to calculate a write incorrectly and overwrite memory beyond the intended allocation. Under the right conditions, that corruption can be developed into information disclosure or code execution.
Java normally manages memory access on behalf of applications. Its runtime, however, also contains low-level components and processor-specific optimizations. This flaw affects one of those components rather than requiring application developers to have written an unsafe memory operation themselves.
The researchers describe successful exploitation in their tests. Those demonstrations establish a security problem, but they do not establish that every organization using the named software has been attacked. A vulnerability, a reachable attack path and an actual compromise remain separate findings.
Hardware and build numbers determine exposure
Calif identifies relevant OpenJDK implementations beginning with version 17.0.6. Reaching the vulnerable optimized code also requires an x86-64 processor with particular capabilities, including AVX-512 VBMI. ARM64 processors do not use that affected path, according to the researchers.
Oracle’s CVE-2026-114938 entry lists 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, along with affected GraalVM releases. That is not a statement that ordinary Java 8 builds or every Java 11 installation are affected by Java4Shell.
The distinction matters when reading an umbrella advisory. A release can need the November update because of another vulnerability even if it does not meet the conditions for this one. OpenJDK’s advisory similarly separates individual CVEs in its risk matrix.
Distribution also matters. Vendors can package fixes under different build identifiers, and enterprise products may bundle their own runtime. Administrators should match the complete build and the application vendor’s guidance rather than rely only on a major-version label such as Java 17.
Identity and development systems are among the tested products
Calif reports demonstrating pre-authentication remote code execution in Keycloak, Atlassian Jira, Confluence, JFrog Artifactory, Jenkins and PingFederate. This is the company’s list of tested products, not a list of confirmed victims or a claim that every deployment is exploitable.
The researchers also emphasize that there is no universal attack that works unchanged against every application. Different targets require different routes to the decoder and different exploitation conditions. The software configuration and how outside data reaches it remain important parts of the assessment.
Because some demonstrated paths occur before authentication, asking users to change passwords does not repair the underlying defect. The central response is to update or mitigate the runtime processing the requests. User-account changes serve a different purpose if a separate investigation finds account compromise.
Oracle rates the issue at 8.1 under CVSS 3.1, while Calif gives it 9.2 under CVSS 4.0. Those are different versions of the scoring framework. Presenting the figures as directly interchangeable would conceal how the assessments were produced.
A patch must reach the running application
New Zealand’s National Cyber Security Centre advises organizations to establish which Java runtime each application actually uses. Its guidance includes runtimes inside containers, copies bundled with products and vendor appliances, rather than only the Java installation visible from an administrator’s shell.
That distinction can determine whether a patch has taken effect. A server may contain more than one runtime, and updating one copy does not establish that every service has switched to it. Applications may need to be restarted, and container images may need to be rebuilt and redeployed.
Calif lists a temporary mitigation using the startup option “-XX:-UseBASE64Intrinsics” to disable hardware-accelerated Base64 operations. For x86-64 systems, it also describes “-XX:UseAVX=2” to disable AVX-512. The latter option is not recognized on ARM and can prevent startup, the company warns.
These measures require testing against the relevant product and environment. Changing a runtime’s processor features can affect performance and operation. A temporary setting should therefore be tracked as part of the response, with the vendor’s corrected release remaining the destination.
Public disclosure also changes what defenders need to distinguish in their reporting. A published laboratory exploit is evidence of capability; it is not evidence that a specific organization lost data. Any conclusion about a breach needs findings from the organization’s systems.
The practical task is to connect the inventory to the live service: identify the application, find its exact runtime and processor, apply the appropriate vendor update and verify what is running afterward. That closes the gap between downloading a patch and actually removing the vulnerable code from use.