# Java4Shell: OpenJDK HotSpot Base64 decoder memory corruption (CVE-2026-114938)

Pack 1.1.0, 19 Nov 2026. CybaVerse threat hunt. Hubfs folder: `openjdk-cve-2026-114938-pack-v1-1-0` (tentative).
This update supersedes pack 1.0.0 (hubfs `openjdk-cve-2026-114938-pack-v1`), which stays unchanged.

Read-only checker (bash and PowerShell), nine KQL hunts, nine Sigma rules. Every hunt is OUR PATTERN: no vendor
publishes indicators of compromise. The only request marker any source states is the 'java4shell-recon' user-agent
(NEWS: SecurityWeek, crediting security researchers), and scanners now also pose as Chrome.

## Changelog

**1.1.0 (19 Nov 2026, sources read 09:35 GMT)**

- Exploitation status: exploit attempts reported by SecurityWeek; no confirmed compromise; not in KEV as of 19 Nov
  2026 (catalogue 2026.11.18). CISA ADP SSVC on the CVE record still says Exploitation "none".
- Scope now confirmed by a vendor: Red Hat states the x86-64 AVX-512 VBMI precondition, that ARM64 is not affected,
  that the code path arrived in 17.0.6 and that OpenJDK 8 and 11 are not affected. Fixed builds unchanged.
- Red Hat rates the CVE CVSS 3.0 9.8 for Red Hat products; the CNA (Oracle) score stays 8.1 High, so the overall
  label stays "Critical by impact (our framing)".
- Downstream: Temurin, Microsoft and Red Hat have now shipped fixed builds.
- New vendor workaround table (Red Hat, Cisco, Elastic, Atlassian, Jenkins, Keycloak, Siemens, CISA), each with the
  vendor's own URL.
- New hunts Q7 (recon user-agent), Q8 (two-stage: recon user-agent, then Chrome from the same IP within 48 hours) and
  Q9 (Base64-heavy parameters, low confidence), with three new Sigma rules. Q1 to Q6 kept.
- Q6 fix: W3CIISLog has no `sComputerName` column; Q6 now uses `Computer`. The KQL parse test now uses the Microsoft
  Learn column lists for W3CIISLog, CommonSecurityLog, Syslog and AzureDiagnostics (`tests/schema/`).
- Checker 1.1.0: reports all four CPU flags Red Hat lists, says "CPU PATH NOT PRESENT, STILL PATCH" for exit 6,
  cites Red Hat for the CPU, 17.0.6, 8 and 11 and workaround notes, and prints the exploitation status line.

**1.0.0 (17 Nov 2026):** first release (hubfs `openjdk-cve-2026-114938-pack-v1`).

## At a glance

| Item | Value | Source |
|---|---|---|
| CVE | CVE-2026-114938, HotSpot compiler (OpenJDK component `hotspot/compiler`; Oracle component "Compiler") | OpenJDK advisory, Oracle CSPU |
| Weakness | CWE-122 Heap-based Buffer Overflow | CISA ADP on the CVE record |
| Vendor score | CVSS 3.1 **8.1 High**, `CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H` (high attack complexity, no authentication) | Oracle (the CNA) on cve.org, the OpenJDK advisory and the Oracle CSPU |
| Researchers' score | CVSS 4.0 9.2, rated critical by the reporting security researchers | NEWS: BleepingComputer |
| Red Hat score | Critical (Red Hat rating) for Red Hat products: CVSS 3.0 9.8, `AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H`. Red Hat is the CNA for its own products only; its page shows cve.org 8.1 | Red Hat CVE page |
| Atlassian score | CVSS 9.2 for Atlassian Data Center products ("the critical risk that this vulnerability represents to customer data") | Atlassian community post (Atlassian Team) |
| Our framing | **Critical by impact (our framing).** The CNA (Oracle) score is 8.1 High, below 9.0, so this pack does not call it Critical overall. Red Hat's 9.8 and Atlassian's 9.2 apply to their own products | OUR ADVICE |
| Exploited | Exploit attempts reported by SecurityWeek; no confirmed compromise; not in KEV as of 19 Nov 2026. CISA ADP SSVC still: Exploitation none, Automatable no, Technical Impact total (record updated 18 Nov 2026 20:51 GMT) | NEWS: SecurityWeek (19 Nov 2026, 09:10 GMT); cve.org (CISA ADP) |
| CISA KEV | Not listed in catalogue 2026.11.18 (released 18 Nov 2026 19:17 UTC, 19:17 GMT), checked 19 Nov 2026 09:35 GMT and rechecked 09:57 GMT. Pack 1.0.0 checked catalogue 2026.11.17 | CISA KEV feed |
| NVD | Published 17 Nov 2026 22:18 UTC; status now Undergoing Analysis (last modified 18 Nov 2026 21:17 UTC); still only Oracle's CVSS 3.1 8.1 (secondary). Pack 1.0.0 saw status Received | NVD |
| Fix | Oracle JDK and OpenJDK 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1 (17 Nov 2026), Oracle Java SE 8u513-perf, and vendor builds below | Oracle JDK release notes, vendor pages |
| Public PoC | BleepingComputer reports a public memory-corruption proof of concept; the app-specific exploits are not public. SecurityWeek: the public proof of concept "only triggers the memory corruption"; one firm has made a full RCE exploit available to its customers and a researcher reports a working exploit, so working RCE exploits exist privately. This pack links no proof of concept | NEWS: BleepingComputer, SecurityWeek |

**Vendor advice (credited to Oracle and OpenJDK):** OpenJDK: "We recommend that you upgrade as soon as possible."
Oracle's Java SE release notes for each 17 Nov build: "This release contains fixes for security vulnerabilities
described in the Oracle Critical Patch Update."

**OUR ADVICE (CybaVerse, not from Oracle):** update every Java 17, 21, 25 and 27 runtime now, internet-facing app
servers first (Jira, Confluence, Jenkins, Keycloak, Artifactory, PingFederate), including JREs bundled inside
products. Where you cannot update today, add `-XX:-UseBASE64Intrinsics` (see "Workaround") and keep the apps behind
SSO or VPN. Patch now regardless of KEV status. Run Q7 and Q8 now: scanners are probing with a 'java4shell-recon'
user-agent and at least one came back posing as Chrome (NEWS: SecurityWeek).

## What each source says about affected and fixed builds

There is no conflict between OpenJDK and Oracle: **both list 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3 and 27.0.1.0.3 as
AFFECTED.** The flag's "fixed after" wording was right about direction, and a later room correction that called
these the fixed builds was wrong (since withdrawn). The fixed builds are the `.1` updates released 17 Nov 2026.

- **OpenJDK advisory 2026/11/17:** "The following vulnerabilities in OpenJDK source code were fixed in this release.
  The affected versions are 27.0.1.0.3, 25.0.5.0.3, 21.0.13.0.3, 17.0.21.0.3, 11.0.33, 8u512, and earlier." That
  sentence covers the whole release (ten CVEs). The per-CVE risk matrix marks CVE-2026-114938 under 17, 21, 25 and
  27 only, not under 8 or 11. So 8 and 11 are out of scope for this CVE (but take their 17 Nov updates for the other
  CVEs).
- **Oracle CSPU Java SE risk matrix and the CVE record (Oracle is the CNA):** "Supported versions that are affected
  are Oracle Java SE: 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3, 27.0.1.0.3; Oracle GraalVM for JDK:
  17.0.21.0.3 and 21.0.13.0.3; Oracle GraalVM Enterprise Edition: 21.3.20.0.3." The record lists every one of these
  with status "affected".
- **Oracle JDK release notes, 17 Nov 2026:** "The full version string for this update release is 17.0.21.1+4",
  and likewise 21.0.13.1+4, 25.0.5.1+4 and 27.0.1.1+4. Each says "This release contains fixes for security
  vulnerabilities described in the Oracle Critical Patch Update." They do not name the CVE.
- **Oracle JDK 8u513-perf release notes:** "The full version string for this update release is 1.8.0_513-perf-b4".
  JDK 8u513 (non-perf) is 1.8.0_513-b04.
- **Oracle GraalVM release notes:** "Oracle GraalVM for JDK 21.0.13.1 (2026-11-17) This is the November 2026 Oracle
  Critical Security Patch Update (CSPU) for Oracle GraalVM for JDK 21. Updated the Oracle JDK release on which Oracle
  GraalVM is built to 21.0.13.1+4." For JDK 17 and for Enterprise Edition the notes conflict, see below.
- **Upstream OpenJDK tags** (as imported by Amazon Corretto): jdk-17.0.21.1+2, jdk-21.0.13.1+2, jdk-25.0.5.1+2 and
  jdk-27.0.1.1+4. Upstream and Oracle builds share the version number; the build number after `+` differs.

**Conflict, stated plainly [REVIEW]:** Oracle's GraalVM release notes say "Oracle GraalVM for JDK 17.0.21.1
(2026-11-17) This is the November 2026 Oracle Critical Security Patch Update (CSPU) for Oracle GraalVM for JDK 17.
Updated the Oracle JDK release on which Oracle GraalVM is built to 17.0.20.1+4." and that Enterprise Edition
21.3.20.1 is built on "17.0.20.1+4 for Java 17-based Oracle GraalVM Enterprise Edition". 17.0.20.1 is below the
Oracle JDK 17 fix (17.0.21.1+4) and below a build Oracle lists as affected. This may be a documentation error, but
until Oracle confirms, the checker reports Oracle GraalVM for JDK 17 at 17.0.21.1 or 17.0.20.1 as REVIEW (exit 3),
never as fixed.

### Affected and fixed table

| Product and line | Affected (per source) | Fixed build | Status of fix |
|---|---|---|---|
| OpenJDK / Oracle JDK 17 | 17.0.21.0.3 and earlier (OpenJDK, Oracle) | 17.0.21.1 (Oracle 17.0.21.1+4) | Shipped 17 Nov 2026 |
| OpenJDK / Oracle JDK 21 | 21.0.13.0.3 and earlier | 21.0.13.1 (Oracle 21.0.13.1+4) | Shipped 17 Nov 2026 |
| OpenJDK / Oracle JDK 25 | 25.0.5.0.3 and earlier | 25.0.5.1 (Oracle 25.0.5.1+4) | Shipped 17 Nov 2026 |
| OpenJDK / Oracle JDK 27 | 27.0.1.0.3 and earlier | 27.0.1.1 (Oracle 27.0.1.1+4) | Shipped 17 Nov 2026 |
| Oracle Java SE 8 Enterprise Performance Pack | 8u511b33-perf (Oracle) | 8u513-perf (1.8.0_513-perf-b4) | Shipped 17 Nov 2026 |
| OpenJDK / Oracle JDK 8 (non-perf) and 11 | Not marked for this CVE; Red Hat: "OpenJDK 8 and OpenJDK 11 do not contain it and are not affected" | n/a (8u513, 11.0.33.1 fix other CVEs) | Out of scope |
| Oracle GraalVM for JDK 21 | 21.0.13.0.3 | 21.0.13.1 (built on 21.0.13.1+4) | Shipped 17 Nov 2026 |
| Oracle GraalVM for JDK 17 | 17.0.21.0.3 | 17.0.21.1 named as the CSPU | [REVIEW] notes say built on 17.0.20.1+4 |
| Oracle GraalVM Enterprise Edition 21.3 | 21.3.20.0.3 | 21.3.20.1 named as the CSPU | [REVIEW] Java 17 base given as 17.0.20.1+4 |
| Amazon Corretto | Earlier builds, for example 17.0.21.0.11, 21.0.13.0.11 | 17.0.21.1.2, 21.0.13.1.2, 25.0.5.1.2, 27.0.1.1.4 | Shipped; each change log lists CVE-2026-114938 (CVSS 8.1, hotspot/compiler) |
| Azul Zulu | Azul lists Zulu 27, 25, 21, 17 as affected | November 2026 CSPU: JDK 27.0.1.1, 25.0.5.1, 21.0.13.1, 17.0.21.1; CPU-line SA builds 25.0.4.1.201, 21.0.12.1.201, 17.0.20.1.201 | Shipped; release notes list CVE-2026-114938 |
| Eclipse Temurin | Earlier 17, 21, 25 and 27 builds | jdk-17.0.21.1+2, jdk-21.0.13.1+2, jdk-25.0.5.1+2, jdk-27.0.1.1+4 (Adoptium API, 19 Nov 2026) | Shipped. Pack 1.0.0 saw jdk-17.0.21+9, jdk-21.0.13+9, jdk-25.0.5+8 and jdk-27.0.1.0.2+1 as latest |
| Microsoft Build of OpenJDK | Earlier 17, 21 and 25 builds | 17.0.21.1, 21.0.13.1, 25.0.5.1 (tags release/jdk-17.0.21.1_2, release/jdk-21.0.13.1_2, release/jdk-25.0.5.1_2) | Shipped: release notes "Last updated on 2026-11-18" point to the OpenJDK advisory; the CVE ID itself is not on the page |
| Red Hat build of OpenJDK (RHEL 8, 9, 10; OpenJDK 17, 21, 25) | Affected per RHSB-2026-018 | For example RHSA-2026:100567 (java-21-openjdk, RHEL 8 and 9) and RHSA-2026:100569 (OpenJDK 21.0.13.1 portable), both issued 18 Nov 2026 | Shipped for those streams; check RHSB-2026-018 for the rest |
| Apps with bundled JREs | See the vendor workaround table | Vendor bundle updates | Check each vendor |

BleepingComputer reports "Java4Shell affects OpenJDK 17.0.6 and later", and Red Hat now states "The vulnerable code
path was introduced in OpenJDK 17.0.6". OpenJDK and Oracle give no lower bound ("and earlier"), so this pack still
treats every 17 build below 17.0.21.1 as affected and the checker notes Red Hat's statement for builds below 17.0.6.

## CPU precondition

- NEWS (BleepingComputer): the flaw "only affects a hardware-accelerated decoding function used on certain x86-64
  processors that support AVX-512 instructions" and needs "supported x86-64 processors with AVX-512 VBMI
  capabilities". OpenJDK and Oracle do not state a CPU condition.
- **Red Hat (primary, CVE page):** "Exploitation requires an x86-64 processor providing the avx512f, avx512bw,
  avx512vbmi, and bmi2 features. Systems running on ARM64 or on x86-64 processors without AVX-512 VBMI support are not
  affected." Cisco: "Hardware families that use x86-64 without AVX-512 or that use different CPU architectures like
  ARM or Cisco Silicon One are not affected." Jenkins, Elastic and Siemens state the same x86-64 AVX-512 VBMI
  condition.
- Checked against the HotSpot source (jdk17u, jdk21u and jdk25u update repositories, read 17 Nov 2026): the x86-64
  Base64 decode stub (`generate_base64_decodeBlock` in `src/hotspot/cpu/x86/stubGenerator_x86_64.cpp`) takes its
  AVX-512 path only when `VM_Version::supports_avx512_vbmi() && VM_Version::supports_avx512bw()`. That code is x86
  only, which matches the vendor statements that ARM64 is not affected (no longer only our assumption).
- Check on Linux (Red Hat RHSB-2026-018): `grep -o avx512vbmi /proc/cpuinfo | head -1` (no output means absent), or
  let either checker read `/proc/cpuinfo` (read only; PowerShell 7 on Linux does the same). Since 1.1.0 both checkers
  also print a `CPU FLAGS (Red Hat list)` line with avx512f, avx512bw, avx512vbmi and bmi2. PRESENT still depends on
  avx512vbmi alone (conservative): a missing bmi2 or avx512f is shown but never turns the result into not affected. On Windows there is no built-in command we have verified for this flag [UNVERIFIED]: save a flag
  list from a hardware inventory tool you trust and pass it with `-CpuFlagsFile`.
- Never treat a missing flag as fixed. A VM guest sees the flags its hypervisor exposes, and a JVM moved to a newer
  host can gain AVX-512 VBMI. The checker therefore reports "AFFECTED, CPU PATH NOT PRESENT, STILL PATCH" (exit 6),
  never "not affected", for an affected version on a CPU without the flag.

## Workaround: exact flag semantics

Both flags were first reported by BleepingComputer and are now in Red Hat RHSB-2026-018, which calls
`-XX:-UseBASE64Intrinsics` "the preferred option, as it removes the vulnerable code path while leaving other
AVX-512 optimizations in place". We checked both in the HotSpot source and in Oracle's `java` manual. They are
temporary; update anyway. Per-vendor advice is in the vendor workaround table below.

| Flag | What the source shows | Effect on this flaw | Cost |
|---|---|---|---|
| `-XX:-UseBASE64Intrinsics` | `product(bool, UseBASE64Intrinsics, false, "Use intrinsics for java.util.Base64")` in `globals.hpp` (17, 21, 25 and mainline). A product flag, not diagnostic or experimental. On x86 the VM turns it on when it is left at default (JDK 17: `UseAVX > 2` with avx512vl and avx512bw; JDK 21 and 25: `UseAVX >= 2`). An explicit `-` keeps it off | Turns off all java.util.Base64 intrinsics, so the AVX-512 decode stub is never used | Base64 falls back to Java code (also loses the AVX2 path on 21 and later) |
| `-XX:UseAVX=2` (or 0 or 1) | `product(int, UseAVX, 3, "Highest supported AVX instructions set on x86/x64")` in `globals_x86.hpp`. With `UseAVX < 3` the VM clears its AVX-512 feature bits, including `CPU_AVX512_VBMI` (`vm_version_x86.cpp`) | `supports_avx512_vbmi()` is then false, so the AVX-512 decode path is not taken | All AVX-512 code paths in the JVM are lost, a wider performance cost |

- **Do you need `-XX:+UnlockDiagnosticVMOptions`?** The HotSpot source says no: `UseBASE64Intrinsics` is declared
  `product`, unlike neighbours such as `UsePoly1305Intrinsics`, which are `DIAGNOSTIC`. Oracle's JDK 21 `java`
  manual, however, says under `-XX:+UseBASE64Intrinsics`: "Flags that control intrinsics now require the option
  -XX:+UnlockDiagnosticVMOptions." OUR ADVICE: put `-XX:+UnlockDiagnosticVMOptions` before the flag. It is harmless
  and satisfies both readings. A JVM that rejects a flag refuses to start, so a running JVM that shows the flag on
  its command line has accepted it.
- **Order matters (OUR ASSUMPTION, standard HotSpot behaviour):** the last setting of a flag wins. A later
  `-XX:+UseBASE64Intrinsics` or `-XX:UseAVX=3` undoes the workaround. The checker reads settings in order.
- **Where to set it:** the app's JVM options file or service definition. `JAVA_TOOL_OPTIONS` also works but is
  read by every JVM on the host. Flags in `@argfiles`, `-XX:Flags=` or `-XX:VMOptionsFile=` files are not read by
  the checker (it warns).
- **`-XX:UseAVX=2` is x86-64 only.** Red Hat: "JVMs on ARM64 do not recognize it and will fail to start if it is
  applied there." Atlassian pairs it with `-XX:+IgnoreUnrecognizedVMOptions` for that reason.
- Confirm on a running JVM with `jcmd <pid> VM.flags -all | grep -F UseBASE64Intrinsics`; Red Hat gives the expected
  output as `bool UseBASE64Intrinsics = false` (the checker does not run it).

## Vendor workaround table (read 19 Nov 2026, 09:35 GMT)

Each row cites the vendor's own page. No row is news-sourced; any vendor row without a primary source would be marked
UNVERIFIED (news-sourced). Atlassian's row comes from a post by the Atlassian Team on the Atlassian Community forum
(signed "Atlassian Support"), not from a security advisory page.

| Vendor | Affected (vendor's own words, shortened) | Fix | Interim workaround | Source |
|---|---|---|---|---|
| Red Hat | RHEL 8, 9 and 10; Red Hat build of OpenJDK 17, 21 and 25 on x86-64 with AVX-512 VBMI | Errata per stream, for example RHSA-2026:100567 and RHSA-2026:100569 (18 Nov 2026) | `-XX:-UseBASE64Intrinsics` (preferred) or `-XX:UseAVX=2` (x86-64 only); verify with `jcmd` | https://access.redhat.com/security/vulnerabilities/RHSB-2026-018 , https://access.redhat.com/security/cve/CVE-2026-114938 |
| Cisco | 13 products affected, including Catalyst Center, Secure Network Analytics, Secure Workload and Intersight Virtual Appliance; IOS XE and UCS Central under investigation (advisory v1.1, severity High) | Fixed releases TBD per Cisco bug | "No workarounds available" | https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-nov2026-java-rce-eCa6ss9z |
| Elastic | Elasticsearch 7.17.3 and later (RCE needs an authenticated user with at least read access; DoS unauthenticated); Logstash 8.6.2 and later only in a specific ingest setup | Upcoming 8.19.26, 9.4.10, 9.5.7 and 9.6.0 with bundled JDK 27.0.1.1+4 (not yet shipped when read) | `-XX:-UseBASE64Intrinsics` in `config/jvm.options.d/`, then restart | https://www.elastic.co/blog/java4shell-remote-code-execution-rce-vulnerability |
| Atlassian | Latest Bamboo, Bitbucket, Confluence, Crowd, Jira and Jira Service Management Data Center in standard installs; scored CVSS 9.2 for Data Center | Out-of-cycle patches once Temurin releases are available | `-XX:UseAVX=2 -XX:+IgnoreUnrecognizedVMOptions` (plus product-specific settings), then restart | https://community.atlassian.com/forums/discussion/3317765/data-center-mitigations-for-cve-2026-114938 |
| Jenkins | Controllers on Java 17, 21 or 25 on x86-64 CPUs with AVX-512 VBMI; ARM64, s390x and ppc64le not at risk | Java 17.0.21.1, 21.0.13.1 or 25.0.5.1, then restart | `-XX:-UseBASE64Intrinsics` on the controller | https://www.jenkins.io/blog/2026/11/18/mitigating-java4shell/ |
| Keycloak | 26.8.2-1 and 26.8.2-2 x86-64 container images (bundled OpenJDK 21.0.13) | Images `latest`, `26.8`, `26.8.2` and `26.8.2-3` ship Red Hat OpenJDK 21.0.13.1 | `JAVA_OPTS_APPEND=-XX:-UseBASE64Intrinsics` | https://www.keycloak.org/2026/11/java4shell-vulnerability-cve-2026-114938 |
| Siemens | Industrial Edge Management Pro V1, Pro V2 and Virtual (SSA-272816 V1.0, CVSS 3.1 8.1) | "Currently no fix is available" | Block direct internet access to IEM Pro and IEM Virtual: "shutdown instances or disable internet connectivity" until a fix is applied | https://cert-portal.siemens.com/productcert/html/ssa-272816.html |
| CISA (alert, not a vendor) | Java installations, including JDKs bundled with third-party software, container images and appliances | Apply vendor patches now, internet-facing first | Restrict and validate Base64 input at the application boundary, enforce input-size limits, limit unauthenticated access; monitor for unexpected JVM crashes and suspicious Java processes | https://www.cisa.gov/news-events/alerts/2026/11/17/critical-java-vulnerability-requires-immediate-patching |

- Elastic narrows the Logstash case to `logstash-filter-elastic_integration` with an ingest pipeline that decodes
  Base64 from untrusted sources, and says the impact there is DoS only. Kibana, Beats and other products without
  bundled Java are not affected.
- Atlassian's post also calls for the Confluence sandbox and Synchrony JVM settings to be updated; use the complete
  product-specific instructions in the post, then restart every relevant service.
- A missing CPU flag is not a patch: "CPU path not present, still patch".

## Exploitation status (NEWS: SecurityWeek, 19 Nov 2026, 09:10 GMT)

Exploit attempts reported by SecurityWeek; no confirmed compromise; not in KEV as of 19 Nov 2026. SecurityWeek, citing
security researchers:

- Attackers are hitting honeypots "with requests decorated in 'java4shell-recon' User-Agents, alongside blind
  exploitation attempts using correctly shaped requests".
- Scanners now pose as Chrome: "an IP address probed a target using a user agent that clearly referenced Java4Shell,
  then returned roughly 17 hours later with the exploit trigger while impersonating the Chrome browser".
- One sensor network "has not seen any successful intrusions on its sensors"; another "cannot confirm whether any
  targets were compromised". Cisco's own advisory: "The Cisco PSIRT is not aware of any malicious use of the
  vulnerability".
- The page's HTML title says "Exploited in the Wild"; its headline and text describe exploitation attempts, which is
  the wording this pack uses.

This pack reproduces no trigger string or payload pattern; none is stated by any source.

## Response steps (OUR ADVICE)

1. Inventory every Java runtime, including bundled JREs and container images (Q1, checker).
2. Update internet-facing x86-64 hosts first; then the rest, ARM64 included, so a later move to an AVX-512 VBMI host
   cannot expose them.
3. Where you cannot update today, apply `-XX:-UseBASE64Intrinsics`, restart and confirm with `jcmd` (see above).
4. Run Q7 and Q8 back to the advisory date. Block or rate-limit client IPs that sent the recon user-agent at your WAF
   or proxy, and review what they requested later under any user-agent.
5. Run Q3 to Q5 and Q9 on hosts those IPs reached; a JVM crash or a new shell from a Java service is a reason to
   start incident response.
6. Recheck the vendor table and KEV daily; fix dates and exploitation status will change.

## Reported impact (NEWS, not verified)

BleepingComputer, citing the reporting security researchers: unauthenticated RCE in default configurations of
Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate; authenticated RCE in
Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit and Spinnaker; exploitation in sample Apache Struts and Spring
Boot applications. "Java4Shell does not have a universal exploit, and attackers must find an application-dependent way
to deliver malicious Base64 data to the vulnerable decoder to achieve code execution." We have not verified any of these app claims. Oracle's
risk matrix note says the flaw "can only be exploited by supplying data to APIs in the specified Component without
using Untrusted Java Web Start applications or Untrusted Java applets, such as through a web service."

## Hunts (all OUR PATTERN)

`kql/java4shell-cve-2026-114938-hunting.kql` (Microsoft Defender XDR and Sentinel) and `sigma/`:

| Hunt | KQL | Sigma | ATT&CK |
|---|---|---|---|
| Java runtime inventory, below the fixed build per line | Q1 (DeviceTvmSoftwareInventory) | none (no Sigma log source for inventory) | T1190 Exploit Public-Facing Application |
| Devices Defender maps to the CVE (empty until Microsoft adds it) [UNVERIFIED] | Q2 (DeviceTvmSoftwareVulnerabilities) | none | T1190 Exploit Public-Facing Application |
| JVM crashes: new `hs_err_pid<pid>.log` | Q3 | `file_event_win_jvm_hs_err_log.yml`, `file_event_lnx_jvm_hs_err_log.yml` | T1499.004 Endpoint Denial of Service: Application or System Exploitation |
| Java app-server process starting sh, bash, dash, zsh, cmd or PowerShell | Q4 | `proc_creation_win_java_app_child_shell.yml`, `proc_creation_lnx_java_app_child_shell.yml` | T1059.001 Command and Scripting Interpreter: PowerShell; T1059.003 Command and Scripting Interpreter: Windows Command Shell; T1059.004 Command and Scripting Interpreter: Unix Shell |
| New public outbound destinations from app-server JVMs | Q5 | `net_connection_win_java_app_outbound.yml` (review list, no baseline) | T1071.001 Application Layer Protocol: Web Protocols |
| Long Base64 runs in web and API logs [UNVERIFIED fields] | Q6 | `webserver_long_base64_request.yml` | T1190 Exploit Public-Facing Application |
| 'java4shell-recon' user-agent in W3CIISLog, CommonSecurityLog, AzureDiagnostics (Application Gateway and Front Door access and WAF logs) and Syslog web access logs (new) | Q7 | `webserver_java4shell_recon_user_agent.yml`, `proxy_java4shell_recon_user_agent.yml` | T1595.002 Active Scanning: Vulnerability Scanning; T1190 Exploit Public-Facing Application |
| Two-stage: recon user-agent from an IP, then a Chrome user-agent from the same IP within 48 hours (new, OUR PATTERN) | Q8 | none (KQL only; needs a join across time) | T1595.002 Active Scanning: Vulnerability Scanning; T1190 Exploit Public-Facing Application |
| Base64-heavy query parameters (new, OUR PATTERN, low confidence) | Q9 | `webserver_base64_heavy_parameters.yml` | T1190 Exploit Public-Facing Application |

- App markers (atlassian, jira, confluence, jenkins, keycloak, artifactory, jfrog, pingfederate) and install-folder
  markers are OUR ASSUMPTION [UNVERIFIED]; adjust them to your hosts.
- The default fatal error log name and location come from Oracle's troubleshooting guide: without `-XX:ErrorFile`
  the default name is `hs_err_pid<pid>.log` (pid is the process ID), written to the working directory of the
  process, or to the temporary directory when that fails.
- Q6: a generic Base64 run of 256 or more characters, plus an `AllAPad` flag for a run of 64 or more "A" characters
  followed by "=". That flag is OUR heuristic, derived from the trigger shape BleepingComputer reports; this pack does
  not reproduce any exploit string. Authorization headers and request bodies are often not logged, so a quiet result
  proves nothing.
- Q7 to Q9 share the `WebRequests` view in the KQL header. W3CIISLog and CommonSecurityLog columns follow Microsoft
  Learn. Application Gateway and Front Door document UserAgent, client IP and request URI fields, but the matching
  AzureDiagnostics column names are [UNVERIFIED], so the KQL uses `column_ifexists`. Their documented WAF log
  schemas have no user-agent field: WAF rows match only when the user-agent appears in the rule match details.
  Syslog parsing assumes the combined access log format [UNVERIFIED]. The web server process list is OUR
  ASSUMPTION.
- Q8: the 48 hour window and the "Chrome/" match are OUR choices, built on the SecurityWeek report of a scanner that
  came back about 17 hours later posing as Chrome.
  Shared, NAT, proxy and CDN addresses join unrelated users; check every hit by hand.
- Q9: two or more parameter values of 64 or more Base64 characters, or Base64 making up 80 percent or more of a
  query of 128 characters or more. OUR thresholds; SAML, OAuth and tracking values give many false positives.
- Sigma conversions for five Kusto pipelines are in `sigma/converted/`, with 17 documented failures (all 9 rules
  convert without a pipeline). Microsoft XDR and Microsoft 365 Defender have no table for the three webserver rules or
  the proxy rule (8). Sentinel ASIM rejects the `cs-uri-query` and `cs-user-agent` fields of the three webserver rules
  (3); it converts the proxy rule. Azure Monitor has no table for the webserver and proxy rules (4) and rejects
  `ParentCommandLine` in the two child-shell rules (2). Use KQL Q6 to Q9 for web logs; see the files.

## Checker

`scripts/check-java4shell.sh` (bash) and `scripts/Check-Java4Shell.ps1` (PowerShell 7) give identical output and
exit codes. They open no sockets, write no files and never run `java`. They read only the files you name, plus
`/proc/cpuinfo` on Linux when no CPU option is given. They print version numbers, vendor names and the flags they
check. `--show-args` prints the JVM arguments with values of `-D` properties and `--options` whose names contain
pass, pwd, secret, token, credential, apikey, api key, private key or auth replaced by `[REDACTED]`, and with the
user and password part of URLs removed.

```
java -version > java-version.txt 2>&1                             # run on the host yourself; the checker never runs java
cp /proc/<pid>/cmdline jvm-args.txt                               # NUL separated; or pass /proc/<pid>/cmdline directly
bash check-java4shell.sh --java-version-file java-version.txt --jvm-args-file jvm-args.txt
bash check-java4shell.sh --java-home "$JAVA_HOME" --cpu-flags-file saved-cpuinfo.txt
pwsh Check-Java4Shell.ps1 -JavaVersionFile java-version.txt -CpuFlagsFile cpu-flags.txt -JvmArgsFile jvm-args.txt
```

Per release line it reports FIXED (at or above 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1), AFFECTED, NOT AFFECTED
(8 non-perf and 11), REVIEW (Oracle GraalVM for JDK 17, see the conflict above) or UNKNOWN (other lines). It
recognises the Oracle 8 perf build (affected below 1.8.0_513-perf) and the Azul Zulu CPU-line builds Azul lists as
fixed. AVX-512 VBMI is PRESENT, ABSENT or UNKNOWN. The workaround is EFFECTIVE, NONE SEEN or UNKNOWN, using the last
setting of each flag.

| Exit | Meaning |
|---|---|
| 0 | NOT AFFECTED: fixed build, or a line OpenJDK and Oracle do not list for this CVE |
| 1 | not used (reserved for unexpected runtime errors) |
| 2 | AFFECTED: affected version, no workaround seen, AVX-512 VBMI present or unknown |
| 3 | UNKNOWN: no usable version, an uncovered line, or REVIEW (sources disagree) |
| 4 | MITIGATED: affected version with an effective workaround flag (temporary) |
| 5 | USAGE OR READ ERROR |
| 6 | AFFECTED, CPU PATH NOT PRESENT (still patch): affected version, no workaround seen, AVX-512 VBMI absent |

Limits: vendor version strings vary; the checker reads the quoted `version "..."` value and the `(build ...)`
values [UNVERIFIED for every vendor]. Bundled JREs need their own `java -version` or `release` file. Environment
variables not in the saved text are not seen.

## Caveats

- The KQL in this pack has NOT been run against a live tenant. It parses cleanly against the Microsoft Learn column
  lists (web tables) and modelled schemas (Defender tables) only.
- Not in CISA KEV (catalogue 2026.11.18, checked 19 Nov 2026 09:35 GMT). BOD 26-04 due dates apply to
  US federal agencies only and are N/A unless CISA adds the CVE. Patch now regardless of KEV status.
- Exploit attempts reported by SecurityWeek; no confirmed compromise; not in KEV as of 19 Nov 2026. CISA ADP still
  records Exploitation "none". A quiet Q7 or Q8 proves nothing: user-agents are trivial to change.
- The OpenJDK and Oracle advisories still state no CPU condition and no 17.0.6 lower bound; Red Hat, Cisco, Elastic,
  Jenkins and Siemens now state the CPU condition, and Red Hat the 17.0.6 start. The app impact lists and the 9.2
  researchers' score remain news claims. We checked the flags and the CPU test in the HotSpot source.
- OUR ASSUMPTION: the last setting of a JVM flag wins.
- Vendor fix status was read on 19 Nov 2026 around 09:35 GMT and will change (Elastic, Atlassian, Cisco and Siemens
  fixes were pending).

## Corrections to the 1.1.0 update flag

1. Red Hat (CNA for its own products) rates the CVE CVSS 3.0 9.8 and errata severity Critical (Red Hat rating) for Red
   Hat products. The CNA (Oracle) score stays 8.1, so "Critical by impact (our framing)" stays for the CVE overall.
2. Scope is confirmed by Red Hat, not only news: x86-64 with avx512f, avx512bw, avx512vbmi and bmi2; ARM64 not affected;
   the code path arrived in 17.0.6; "OpenJDK 8 and OpenJDK 11 do not contain it". The fixed builds stand.
3. NVD status moved from Received to Undergoing Analysis. KEV is now catalogue 2026.11.18 (CVE not listed). CISA
   ADP SSVC still says Exploitation "none".
4. Temurin, Microsoft and Red Hat have now shipped fixed builds, not only Corretto and Azul.
5. Cisco lists 13 affected products (plus two under investigation), rates its advisory High and says "No workarounds
   available". Elastic fixes are announced, not shipped. Atlassian's mitigation is `-XX:UseAVX=2` with
   `-XX:+IgnoreUnrecognizedVMOptions`, from a vendor community post, with a CVSS 9.2 product score. Keycloak has fixed
   image tags. Siemens has no fix yet.
6. The recon user-agent and the Chrome return are reported by named security firms quoted by SecurityWeek; this pack
   credits security researchers only. SecurityWeek says the public proof of concept only triggers the memory corruption; full
   RCE exploits are held privately (one offered to a firm's customers), not public.

## Corrections to the original flag (1.0.0)

1. "Fixed after 17.0.21.0.3, ..." is right in direction: those builds are the last affected (OpenJDK: "The affected
   versions are ..."; Oracle: "Supported versions that are affected are ..."). The fixed builds are 17.0.21.1,
   21.0.13.1, 25.0.5.1 and 27.0.1.1 (Oracle +4). The room correction calling the `.0.3` builds fixed was wrong.
2. Oracle also lists Oracle Java SE 8u511b33-perf (fixed in 8u513-perf), Oracle GraalVM for JDK 17.0.21.0.3 and
   21.0.13.0.3, and GraalVM Enterprise Edition 21.3.20.0.3. OpenJDK 8 and 11 are not marked for this CVE; the
   advisory's "11.0.33, 8u512" applies to the release as a whole.
3. GraalVM for JDK 17 (17.0.21.1) and GraalVM EE 21.3.20.1 release notes give a Java 17 base of 17.0.20.1+4, which
   conflicts with the fix; marked REVIEW.
4. The AVX-512 VBMI precondition, the 17.0.6 start, the app lists, the 9.2 score and both flags are news claims; the
   advisories do not state them. The HotSpot source confirms the AVX-512 decode path needs avx512vbmi and avx512bw.
5. `UseBASE64Intrinsics` is a product flag in the HotSpot source and needs no unlock there; Oracle's `java` manual
   says intrinsic flags need `-XX:+UnlockDiagnosticVMOptions`. Use both. `-XX:UseAVX=2` works because `UseAVX < 3`
   clears all AVX-512 features, at a wider performance cost.
6. CNA is Oracle (cve.org), CVSS 3.1 8.1 High; CISA ADP adds CWE-122 and SSVC Exploitation none. NVD status
   Received. Not in KEV.
7. Downstream: Corretto and Azul Zulu have shipped fixes that name the CVE; Temurin and Microsoft had not shipped
   when checked; Red Hat had no CVE page. BleepingComputer's list of affected distributions is a news claim.

## Credits

Research is credited to security researchers only, as reported by BleepingComputer and SecurityWeek. This pack names
no individual or firm from the OpenJDK, Oracle or vendor acknowledgements or from the news reports.

## Sources

- OpenJDK Vulnerability Advisory 2026/11/17: https://openjdk.org/groups/vulnerability/advisories/2026-11-17
- Oracle Critical Security Patch Update, November 2026: https://www.oracle.com/security-alerts/cspunov2026.html
- CVE record: https://www.cve.org/CVERecord?id=CVE-2026-114938
- NVD: https://nvd.nist.gov/vuln/detail/CVE-2026-114938
- CISA KEV catalogue: https://www.cisa.gov/known-exploited-vulnerabilities-catalog
- Oracle JDK release notes: https://www.oracle.com/java/technologies/javase/17-0-21-1-relnotes.html ,
  https://www.oracle.com/java/technologies/javase/21-0-13-1-relnotes.html ,
  https://www.oracle.com/java/technologies/javase/25-0-5-1-relnotes.html ,
  https://www.oracle.com/java/technologies/javase/27-0-1-1-relnotes.html ,
  https://www.oracle.com/java/technologies/javase/8u513-perf-relnotes.html
- Oracle GraalVM release notes: https://docs.oracle.com/en/graalvm/jdk/17/docs/release-notes/ ,
  https://docs.oracle.com/en/graalvm/jdk/21/docs/release-notes/ ,
  https://docs.oracle.com/en/graalvm/enterprise/21/docs/release-notes/
- Oracle `java` manual (JDK 21): https://docs.oracle.com/en/java/javase/21/docs/specs/man/java.html
- Oracle fatal error log: https://docs.oracle.com/en/java/javase/21/troubleshoot/fatal-error-log.html
- Azul Zulu November 2026 release notes: https://docs.azul.com/core/release/november-2026/release-notes
- Adoptium release list: https://api.adoptium.net/v3/info/release_names
- Microsoft Build of OpenJDK release notes: https://learn.microsoft.com/en-us/java/openjdk/release-notes
- Amazon Corretto 17, 21, 25 and 27 change logs (Amazon's Corretto source repositories; not linked here)
- HotSpot source, jdk17u, jdk21u and jdk25u update repositories (files named above; not linked here)
- Red Hat: https://access.redhat.com/security/cve/CVE-2026-114938 ,
  https://access.redhat.com/security/vulnerabilities/RHSB-2026-018 , https://access.redhat.com/errata/RHSA-2026:100567 ,
  https://access.redhat.com/errata/RHSA-2026:100569
- Cisco: https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-nov2026-java-rce-eCa6ss9z
- Elastic: https://www.elastic.co/blog/java4shell-remote-code-execution-rce-vulnerability
- Atlassian (community post by the Atlassian Team): https://community.atlassian.com/forums/discussion/3317765/data-center-mitigations-for-cve-2026-114938
- Jenkins: https://www.jenkins.io/blog/2026/11/18/mitigating-java4shell/
- Keycloak: https://www.keycloak.org/2026/11/java4shell-vulnerability-cve-2026-114938
- Siemens SSA-272816: https://cert-portal.siemens.com/productcert/html/ssa-272816.html
- CISA alert, 17 Nov 2026: https://www.cisa.gov/news-events/alerts/2026/11/17/critical-java-vulnerability-requires-immediate-patching
- Microsoft Learn table references: https://learn.microsoft.com/en-us/azure/azure-monitor/reference/tables/w3ciislog ,
  https://learn.microsoft.com/en-us/azure/azure-monitor/reference/tables/commonsecuritylog ,
  https://learn.microsoft.com/en-us/azure/azure-monitor/reference/tables/syslog ,
  https://learn.microsoft.com/en-us/azure/azure-monitor/reference/tables/azurediagnostics ,
  https://learn.microsoft.com/en-us/azure/application-gateway/monitor-application-gateway-reference ,
  https://learn.microsoft.com/en-us/azure/frontdoor/front-door-diagnostics
- NEWS: https://www.bleepingcomputer.com/news/security/critical-java4shell-flaw-allows-rce-attacks-on-major-java-applications/
- NEWS: https://www.securityweek.com/java4shell-exploitation-attempts-begin-as-vendors-rush-out-advisories/
- MITRE ATT&CK: https://attack.mitre.org/techniques/T1190/ , https://attack.mitre.org/techniques/T1499/004/ ,
  https://attack.mitre.org/techniques/T1071/001/ , https://attack.mitre.org/techniques/T1595/002/

## Files

`scripts/` checkers; `kql/` hunts; `sigma/` rules and `sigma/converted/`; `tests/` (bash and Pester tests, shared
cases, fixtures marked FIXTURE_ONLY, static and negative gates, `run_all.sh`).
