First-Day Reports Rated Java4Shell 9.2; Its CVE Record Rates It 8.1
Java4Shell, also called Java2Shell, is a heap overflow in the Base64 decoder of OpenJDK on x86-64. We read the ten TLP:CLEAR reports in Asterism's panel from its first day, 17 Nov 2026, and checked each claim against the researchers' write-up and the CVE record.

Java4Shell (CVE-2026-114938) was disclosed on 17 Nov 2026, and by the end of the day 10 TLP:CLEAR reports in Asterism's panel covered it. Of those, 7 cite a severity of 9.2. The CVE record that Oracle published at 21:32 UTC scores it 8.1 and rates it difficult to exploit. The reports also split on which Java versions are affected and on whether a fix had shipped. They agreed on what had not happened: no exploitation reported, and no public exploit.
Key Findings
- Of the 10 reports, 7 cite 9.2 under CVSS v4.0. The CVE record gives 8.1 under CVSS v3.1, with high attack complexity. Both scores appear in 6 reports.
- On the fix, 3 reports tie the CVE to Oracle's update of 17 Nov, and 6 say no fixed build was named or mapped to it yet. The difference is mostly timing: the posts that hedge describe Oracle's pre-release notice, the others its published update.
- The reports name three different sets of affected versions: Java 17; Java 17, 21 and 25; or Java 17, 21, 25 and 27. The CVE record lists 5 Oracle Java SE builds as affected, from
8u511b33-perfto27.0.1.0.3. - No report describes exploitation in the wild, and 5 say plainly that none was reported. CISA's KEV catalog doesn't list the CVE. The only public code is the researchers' reproducer, which triggers the bug; their product exploits are withheld.
- The researchers named it Java4Shell, and 8 reports use that name; 2 use only the CVE. "Java2Shell" appears in none of them: it comes from posts on X and a vendor's site.
The Flaw, According to Its Sources
Calif, the researchers who found the flaw, describe a heap out-of-bounds write in the HotSpot JVM's hardware-accelerated Base64 decoder, the code path used on x86-64 processors with AVX-512 VBMI.1 ARM servers don't take that path. Calif says it demonstrated unauthenticated code execution in the default configurations of six products (Keycloak, Jira, Confluence, Artifactory, Jenkins and PingFederate), that an AI coding agent found the bug in about an hour, and that Oracle patched it on 17 Nov. It published a reproducer that triggers the overflow and withheld the product exploits. The other name, Java2Shell, comes from posts on X and a guide run by WiseBee, a security vendor.2
Oracle, as the CVE Numbering Authority, published the record the same evening: an issue in the Compiler component of Oracle Java SE and GraalVM that an unauthenticated attacker can exploit over the network, though with difficulty.3 CISA's alert urges patching, and while that isn't possible, restricting and validating Base64 input at the application's edge.4 Three reports name the same two JVM options that turn the vulnerable code path off, -XX:-UseBASE64Intrinsics and -XX:UseAVX=2; VulnCheck says it checked them on Java 17, 21 and 25.5 We didn't test them.
Where the First Day Hedged
as of 17 Nov 2026Measures reporting, not activity.small sample (n = 10)Download CSV
Data table hedged.csv
| question | reports_hedging |
|---|---|
| affected versions | 4 |
| pre-auth code execution | 5 |
| severity score | 5 |
| a fix | 6 |
| public exploit code | 0 |
| exploitation | 0 |
| how hard to exploit | 1 |
Across the seven questions, the reports hedged 21 times and spoke plainly 26 times. The hedges cluster on what the vendors had not yet published: which versions, which score, which build fixes it. The plain statements cluster on what had not happened: 5 reports say no exploitation was reported, and 4 that no exploit is public beyond the reproducer; none hedged on either. Equixly's report is the one that disputes a claim: it calls the code execution unproven, and a crash the realistic outcome.6
The Record Against the Reports
| question | the record | the reports |
|---|---|---|
| severity | 8.1, CVSS v3.1, high complexity | 7 cite 9.2, CVSS v4.0 |
| affected | 5 Java SE builds, 8u511b33-perf to 27.0.1.0.3 | three different sets |
| fix | Oracle's update of 17 Nov | 3 tie it, 6 hedge |
| exploitation | not in KEV | none reported |
| name | Java4Shell | none use Java2Shell |
The two scores need not conflict. They come from different versions of CVSS and different scorers, and Oracle's vector rates the attack complexity high where the early 9.2 did not. Aikido and Equixly credit the 9.2 to an advisory from South Korea's NCSC dated 16 Nov, which none of the reports link.7 Three reports say details leaked or were teased before the patch.
The affected versions differ for a similar reason. Oracle's record lists the latest build of each supported line it found affected, which covers the earlier builds of that line; the earlier reports named what had been confirmed publicly when they wrote. Upwind's report, updated after the advisories, names Oracle's fixed build for each of Java 17, 21, 25 and 27, from Oracle's advisory (the build numbers are in Upwind's report: written out, they read like IP addresses, and our indicator checks would rightly stop them).89 Tenable's says plainly that Oracle patched the flaw in its update of 17 Nov.10
Limits of This Report
- This is day zero. Reports were still arriving while we read: a tenth came in during the coding. Changes will come as a dated update, never a silent edit.
- This measures reporting, not activity. "No exploitation reported" is what the reports say, not something we observed.
- Our panel is mostly English-language security vendors, with CISA. Korean-language sources, including the NCSC advisory and the press reports that broke the story there, are outside it, and so are posts on social media.
- Oracle's and OpenJDK's advisory pages refused our automated reads. What they say here comes from the CVE record and from the reports that quote them.
- Ten reports are a small sample. We show counts, not shares with intervals.
- We didn't test any mitigation or reproduce the flaw.
Method
The definitions of a report and an organization are those of How Null Sec Counts. The set is every TLP:CLEAR report in Asterism's panel dated 17 Nov 2026 whose text names Java4Shell, Java2Shell or the CVE: 10 reports from 10 organizations, 9 of them vendors. We read each one in full and coded what it says about seven questions as said plainly, hedged (reported as unconfirmed or as someone else's claim), disputed, or not addressed, with the source it credits. The questions were fixed before the reading. The coding is in reports.csv; the CVE record's fields and the KEV and EPSS checks are in record.csv, fetched on 17 Nov, and every query is in queries.aql.
About the Data
- As of
- Window
- 17 Nov to 17 Nov 2026
- Coverage since
- Panel
- 64 sources reported that day; the data counted here come from 10 organizations
- Marking
- TLP:CLEAR
- Left out
- Undated reports, marketing and product news, publishers who opted out of AI use, anything above TLP:CLEAR, and posts on social media.
This measures reporting, not activity. It counts what was published.
Null Sec sells Asterism, the platform these data come from.
Data and Queries
The files behind every figure and number in this report, as published. Each SHA-256 lets you check you have the same bytes.
- hedged.csvThe data behind Fig. 1.155 bytes · SHA-256
195f6c83f46190b56c3628ab39631513929e9ddcb2449d741775581b5b007292 - record.csvSupporting data.857 bytes · SHA-256
3072c5d878bad78ead895021dcc03bfe8f484b95fc9c3c6a18e5da3a0ffb7430 - reports.csvSupporting data.5.5 KB · SHA-256
15d20c46978d7db3811ed46d238dedbcf17284b2853dd885e9e9d6492276bb32 - scores.csvSupporting data.50 bytes · SHA-256
5fa92b3b32019edfcfa2d91a31cc3c68706676971bb3fa22c6394f4266974db2 - stances.csvSupporting data.405 bytes · SHA-256
508ff5ffd6987f54178490c4304b9ddaf5b1719fee76c29a5862baba09e0317f - queries.aqlThe exact queries, with their windows and the dates they ran.1,000 bytes · SHA-256
363c63ce5e95615f3f9ec88bf4434840bf132209390c5c558898133c7a0aeea9 - sources.csvEvery report cited or counted, with its publisher, date, URL and TLP.3.2 KB · SHA-256
239655ea77c7e864e09cda6c78bba3c0b4bed5e34635e9769997f7b4c5097c9d - public.csvThe public datasets used, with the dates we fetched them.466 bytes · SHA-256
4ec5a6f7d4c1163b021dbeaeb74d6a65105aef2de0b902781e79623af6792503
References
- Calif, "Java4Shell", 17 Nov 2026. calif.io/java4shell/ ↑
- WiseBee, "Java2Shell", 17 Nov 2026: the source of the alternative name. java2shell.io/ ↑
- CVE Program, CVE-2026-114938, published 17 Nov 2026 by Oracle as the CNA. www.cve.org/CVERecord?id=CVE-2026-114938 ↑
- CISA, "Critical Java Vulnerability Requires Immediate Patching", 17 Nov 2026. www.cisa.gov/news-events/alerts/2026/11/17/critical-java-vulnerability-requires-immediate-patching ↑
- VulnCheck, "Java4Shell CVE-2026-114938: Fine, I Guess We'll Do This Again", 17 Nov 2026. www.vulncheck.com/blog/java4shell ↑
- Equixly, "CVE-2026-114938: Pre-auth heap overflow in Java 17, 21, and 25", 17 Nov 2026. equixly.com/blog/2026/11/17/cve-2026-114938-pre-auth-heap-overflow-in-java-17-21-and-25/ ↑
- Aikido Security, "Pre-auth RCE flaw in Java 17 puts servers at risk: affected versions and how to fix (CVE-2026-114938)", 17 Nov 2026. www.aikido.dev/blog/java-17-rce-vulnerability-cve-2026-114938 ↑
- Upwind, "Java4Shell: What We Know Now and What Comes Next", 17 Nov 2026, updated after the advisories. www.upwind.io/blog/java4shell-what-we-know-now-and-what-comes-next ↑
- Oracle, "Critical Security Patch Update Advisory, November 2026", 17 Nov 2026. www.oracle.com/security-alerts/cspunov2026.html ↑
- Tenable, "Oracle November 2026 Critical Security Patch Update Addresses 640 CVEs (CVE-2026-114938)", 17 Nov 2026. www.tenable.com/blog/oracle-november-2026-critical-security-patch-update-addresses-640-cves-cve-2026-114938 ↑
Spot an error? Write to us and we'll fix it here, with a dated note.