Oracle fixed CVE-2026-114938 on November 17, 2026, according to SecurityOnline, but that date does not tell an OpenJDK operator when their own runtime is patched. The useful unit of remediation is not the Java major version. It is the exact vendor build running inside each service, container, CI image or operating-system package.
Amazon Corretto shipped a fixed build the same day, 17.0.21.1.2, while an Aikido tracker snapshot showed “No confirmed fix” for Eclipse Temurin, a status ThreatDefence’s November 18 table echoes with “No fixed build yet.” That phrase is absence of confirmation, not proof that Temurin has failed to ship. The other channels are not undocumented either: the same table reports Red Hat’s fixes expedited with its fixed builds still pending, and describes Azul Zulu, Microsoft, BellSoft, SAP and Linux distributions as “Built from the same code and affected,” pointing to each vendor’s advisory for the fixed build.
The practical answer is therefore uncomfortable but actionable: enumerate each runtime’s provenance now, compare its full build with that distributor’s own advisory, and keep remediation open where no fixed threshold can be confirmed. Switching channels solely because Oracle has patched would exchange one unverified assumption for another.
What shipped on November 17, and what remains unclear
Two reports independently date the fix to November 17. ThreatDefence attributes it to Oracle and the OpenJDK Vulnerability Group, while SecurityOnline places it in Oracle’s Critical Security Patch Update. That establishes Oracle’s release event, not the contents or availability of every downstream package.
The Oracle fixed builds themselves are reported consistently by two secondary sources: ThreatDefence lists 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1 from the November CPU, and Aikido’s table carries the same four as fixed versions. Neither quotes Oracle’s own advisory, so confirm those identifiers against Oracle’s release notes before using them as fleet thresholds, just as each other distributor’s advisory remains necessary for its packages.
The technical picture is also unsettled. SecurityOnline reports that Calif attributes Java4Shell to memory corruption in OpenJDK’s Base64 decoder and says it enabled pre-authentication remote code execution in Jira, Confluence, Jenkins and Keycloak. That is a reported root cause and set of affected deployments, not evidence that every installation of those products is exploitable.
Meanwhile, Aikido’s tracker said at publication that the vulnerable component and exploitation conditions were not public, with further details expected after Oracle’s release. The pages may simply reflect different publication times, but these snapshots do not settle which account is current. Until primary advisories and technical details settle that disagreement, exposure rules and interim mitigations need to remain conditional.
Why “patched” depends on where the JDK came from
Aikido states that OpenJDK vendors ship on their own schedules. Consequently, the upstream fix date and a downstream installation date are different facts. A service can already contain the fix, await a distributor build, or run an image that has not been rebuilt even though its maintainer has published one.
Version strings need equally careful handling. Corretto’s documented identifiers do not have the same shape as Temurin’s, and a build number from one distribution cannot be applied to another. Even matching the numeric Java release is insufficient if the vendor suffix or package revision identifies a different patch lineage.
The Temurin entry illustrates the distinction between status and evidence. Aikido’s November 18 snapshot said “No confirmed fix,” and ThreatDefence’s table, also dated November 18, reports “No fixed build yet,” with the current 17.0.21, 21.0.13 and 25.0.5 builds affected. Those are same-day snapshots of a moving target, not a shipping forecast: keep checking Adoptium’s advisory and release notes rather than treating a different vendor’s build as a drop-in security equivalent.
The runtime source to patch channel matrix
This matrix separates documented findings from verification work. All version ranges, build thresholds and vendor statuses are self-reported snapshots as of November 18, 2026, not independently validated package tests.
| Runtime source | Reported affected scope | Documented patched threshold | Operational status and verification channel |
|---|---|---|---|
| Oracle Java SE | Semgrep’s self-reported Oracle list includes 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3 and 27.0.1.0.3 (report). | 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1, reported by both ThreatDefence and Aikido. | Reported by two secondary sources, but no primary Oracle advisory is quoted by either. Confirm the build and product line against the Oracle CPU and release notes before fleet-wide use. |
| Oracle OpenJDK and Oracle GraalVM for JDK | Both are named in Semgrep’s affected-distribution list; ThreatDefence additionally reports Oracle GraalVM for JDK 17.0.21.0.3 and 21.0.13.0.3, and GraalVM Enterprise Edition 21.3.20.0.3, as affected. | Not established. | Check the corresponding Oracle OpenJDK or GraalVM advisory rather than borrowing Java SE thresholds. |
| Amazon Corretto | Aikido’s self-reported ceilings are Java 17 through 17.0.21.0.11, Java 21 through 21.0.13.0.11, Java 25 through 25.0.5.0.10 and Java 27 through 27.0.1.0.11 (tracker). | Aikido lists self-reported fixed builds 17.0.21.1.2, 21.0.13.1.2, 25.0.5.1.2 and 27.0.1.1.4 (tracker). | These are the clearest downstream thresholds in this reporting. Confirm the installed build through Corretto’s advisory or release notes. |
| Eclipse Temurin | Aikido’s self-reported ranges are Java 17 through 17.0.21+9, Java 21 through 21.0.13 and Java 25 through 25.0.5+8 (tracker). | No fixed build as of November 18: “No confirmed fix” in Aikido’s snapshot, “No fixed build yet” per ThreatDefence. | Recheck Adoptium’s advisory and release notes, and consider the interim workaround below. A same-day snapshot is not a shipping forecast. |
| Microsoft Build of OpenJDK | Named by Semgrep, without a version range; ThreatDefence groups it among vendors “Built from the same code and affected.” | Not established. | Check Microsoft’s own security advisory and release notes for its fixed build. |
| Azul Zulu | Named by Semgrep, without a version range; “Built from the same code and affected” per ThreatDefence. | Not established. | Check Azul’s advisory and the exact Zulu package build. |
| BellSoft Liberica | Named by Semgrep, without a version range; “Built from the same code and affected” per ThreatDefence. | Not established. | Check BellSoft’s advisory and Liberica release notes. |
| SAP SapMachine | Named by Semgrep, without a version range; “Built from the same code and affected” per ThreatDefence. | Not established. | Check SAP’s advisory and SapMachine release notes. |
| Red Hat OpenJDK | Named by Semgrep; ThreatDefence reports it affected on RHEL 8, 9 and 10 for OpenJDK 17, 21 and 25. | Fixed build pending as of November 18 (ThreatDefence). | Red Hat published bulletin RHSB-2026-018 on November 17 and says fixes are being expedited. Check the advisory and the installed package revision. |
| Linux distro package or container base image | Reported “Built from the same code and affected” by ThreatDefence; no universal range exists independently of the bundled JDK. | Determined by the packager or image maintainer’s released runtime. | Identify the bundled vendor and build, then check both the package or image advisory and the runtime actually executed. This row is an operational inference from provenance, not a per-distro verified status. |
For Corretto, compare the complete build against the appropriate row, not merely the Java 17, 21, 25 or 27 prefix. For channels whose fixed build is pending or has not been named, the correct state is unresolved, not “patched” and not automatically “vulnerable.” Unresolved means the remediation ticket still needs evidence.
Turning java -version into a fleet decision
A useful inventory can be built without first knowing the vulnerable component. The procedure is to establish provenance, then attach vendor evidence to each exact runtime.
-
Run
java -versionin the environment that executes the service. A workstation JDK does not establish what runs in a container, CI job or server package. For containers, check the running workload and retain the image identity so the result can be reproduced. -
Record the vendor and the complete version string. Preserve suffixes and package revisions. Where the output does not make provenance clear, use the package or image metadata as a second identifier rather than guessing from the major version.
-
Map that provenance to one patch channel. An Oracle threshold answers an Oracle question. A Corretto threshold answers a Corretto question. A base image adds another question: whether the image consumed by the service actually contains its maintainer’s updated runtime.
-
Attach an advisory, threshold and check time to every row. Useful statuses are patched with a cited fixed build, affected below a documented fixed build, or unresolved because the relevant advisory or threshold has not been confirmed.
-
Re-run the check after remediation. Updating a package manifest or pulling an image tag is not, by itself, evidence that the restarted service executes the fixed runtime. The post-change
java -versionresult closes that loop.
For CI images, this enumeration should include build and test runners, not only production deployments. The reporting does not establish that a CI runtime has the same exposure as an internet-facing service, but recording it prevents an old JDK from surviving unnoticed in a rebuild pipeline and later re-entering an artifact or environment.
The cost here is real inventory work, especially where base images have accumulated without clear ownership. Still, that work is more durable than a one-off version sweep: the next OpenJDK security release will depend on the same provenance records.
The affected-version boundary is contested
Aikido’s self-reported branch list covers Java 17, 21, 25 and 27 before their patched builds. Semgrep instead reports OpenJDK 17.0.6 and later, describing that as every OpenJDK release since 2021. Its wording is broader than four branches and would include intervening feature releases.
Semgrep’s self-reported Oracle build list also includes Java SE 8u511b33-perf (same report). That directly conflicts with an assumption that Java SE 8 is categorically unaffected; the affected item is the Oracle Java 8 Performance Pack specifically. On Java 11, the reporting agrees rather than conflicts: Aikido writes “Java SE 8 and 11 are not affected by this CVE,” ThreatDefence reports “Standard Java 8 and 11 are not affected,” and SecurityOnline notes “Standard Java 8 and Java 11 builds are not listed for this flaw.”
This changes the inventory rule. I would not filter the first pass to Java 17, 21, 25 and 27 alone. Include OpenJDK runtimes within Semgrep’s reported 17.0.6-and-later scope and flag the reported Oracle Java 8 Performance Pack build. Standard Java 11 is reported unaffected by all three analysis sources, so it can sit outside the urgent queue while staying inside ordinary patch hygiene; recording it during enumeration costs little if you would rather not lean on a single day’s snapshots. That is a triage choice, not a declaration that every included runtime is exploitable.
Severity helps prioritize, but it does not close the exposure question
The two severity figures are both self-reported: Calif’s 9.2 under CVSS 4.0 and Oracle’s 8.1 under CVSS 3.1, according to SecurityOnline. They come from different reporters and different CVSS versions. Neither number identifies which local services meet the exploitation conditions.
SecurityOnline also carries the self-reported status that no exploitation in the wild had been reported and that Calif had not released its exploits. That is useful context for communication, not an assurance of continued safety or evidence that exploitation cannot already be occurring unnoticed.
The reports do agree on one concrete exposure boundary: the vulnerable assembly runs only on x86-64 processors with AVX-512 VBMI. They attribute it differently, though. ThreatDefence credits Calif with the boundary, while SecurityOnline cites Equixly’s analysis for the qualifying CPUs, “Intel Ice Lake or later and AMD Zen 4 or later,” and adds that “ARM hosts are not affected.” Semgrep states the same hardware requirement without naming Equixly. That is self-reported analysis, not a validated rule, but it means two services running the identical Java build can differ in exposure by CPU. Both Semgrep and ThreatDefence still recommend updating affected versions regardless of architecture, because exposure is hard to pin down and workloads move between instance types.
Use that boundary to order the queue, not to certify hosts: prioritize services with a documented affected runtime on qualifying x86-64 processors, especially internet-facing ones, but do not clear the rest from version data alone. For channels with no fixed build yet, ThreatDefence reports an interim JVM workaround that disables the accelerated Base64 decoder with -XX:-UseBASE64Intrinsics; SecurityOnline notes that neither Oracle nor OpenJDK has confirmed the workaround, so test it before relying on it.
The decision rule is consequently narrow and defensible: deploy a distributor-documented fixed build through the channel that supplied the runtime, verify the executing version afterward, and keep every unresolved provenance row visible. Oracle’s November 17 release started that process; it did not finish it for the rest of the fleet.
Frequently Asked Questions
What is the interim JVM workaround for Java4Shell if a fixed build is not yet available?
For channels with no fixed build yet, ThreatDefence reports an interim JVM workaround that disables the accelerated Base64 decoder with -XX:-UseBASE64Intrinsics; SecurityOnline notes that neither Oracle nor OpenJDK has confirmed the workaround, so test it before relying on it.
Which CPU architectures are affected by the Java4Shell vulnerability?
The reports do agree on one concrete exposure boundary: the vulnerable assembly runs only on x86-64 processors with AVX-512 VBMI. They attribute it differently, though. ThreatDefence credits Calif with the boundary, while SecurityOnline cites Equixly’s analysis for the qualifying CPUs, “Intel Ice Lake or later and AMD Zen 4 or later,” and adds that “ARM hosts are not affected.”

Join the discussion
Share a useful perspective or ask a question about this article.