Java4Shell: GreyNoise 187 sessions, no confirmed success; Previdian 9 attempts; Cisco, Siemens list no fix
As read at about 12:15 UTC on 19 November 2026, GreyNoise's tracker shows five sources and 187 Java4Shell shaped sessions with no confirmed success, Previdian reports 9 attempts, and Cisco and Siemens list no fix. The counts measure different things over different windows.
By Parminder Kumar Sharma · LinkedIn · · 27 min read
GreyNoise counts 187 sessions from five sources, and none confirmed successful
As this site read GreyNoise's Java4Shell tracker at about 12:15 UTC on 19 November 2026, it listed five source addresses that had sent a Base64 value of the Java4Shell trigger shape in an HTTP Authorization header since 00:00 UTC on 17 November. The page prints their session counts as 70, 64, 25, 22 and 6, which add up to 187 (derived: 70 + 64 + 25 + 22 + 6). The same page prints "Successful Exploitation (0)" and says "No confirmed successful exploitation against GreyNoise gridpoints yet." Java4Shell is CVE-2026-114938, the OpenJDK Base64 decoder memory corruption covered in brief 733.
What that does not establish. A session at one company's sensors whose header had a certain shape is not a working exploit, not proof that any target was vulnerable, and not a measure of what happened anywhere else. The zero is also one network's view: it says nothing about servers that network does not watch. The saved pages print no indicators of successful compromise. This site's copy of the tracker had its address list, network owners, hashes and paths removed before it was kept, so no address and no network owner is printed here.
Other counters appeared over the same two days. Previdian's page reports 9 exploitation attempts from 6 addresses in 4 countries since 18 November, and SecurityWeek, a secondary report published at 09:10:58 UTC on 19 November, relays statements from WatchTowr, Cloudflare, VulnCheck, Immersive and Cisco. They count different things over different windows, so this brief does not add them. This site read the GreyNoise, Previdian and Cisco pages itself. The other four sources' own pages were not linked or sit behind a login, so each is attributed to SecurityWeek and was not read here.
What changed since brief 733. Brief 733 said no saved page then reported exploitation; that was true of the four pages it read on 18 November between about 07:10 and 07:40 UTC. The pages read today report attempts and no confirmed success, and SecurityWeek relays a report that a full exploit is in customers' hands. The table in the section on what the pages do not state sets the two readings side by side.
Who is selling what. GreyNoise and Previdian sell threat intelligence, and the Previdian page carries "Request a Demo" buttons beside its figures. Azul sells Java builds and cites its own survey in its post. Cisco, Siemens, Elastic, Red Hat, Jenkins, Keycloak and Atlassian describe their own products and are attributed as such.
Who counts what: five counters measuring five different things
The figure sets each page's own wording side by side. It draws no bars, because a session, an attempt, a honeypot request and a returning address are not the same unit, and a bar would invite a comparison the pages do not support.
GreyNoise. The page defines its five as source addresses whose Authorization header carried a Basic or Bearer value of the trigger shape, which it describes as 64 or 128 characters plus 1 to 3 more, ending in an equals sign, with benign sources excluded. It prints other counts under other definitions: 12 "block candidates" (tracked addresses classified malicious or suspicious, excluding BSI trust levels 1 and 2), one newly observed attack infrastructure entry, and two similarity clusters that it calls investigative leads, not attribution. This site's copy does not say how the 12 relate to the five, so it does not equate them.
Previdian. The page reports 9 attempts, 6 unique attacker addresses, 4 countries (CN, DE, RO and US) and 5 sensors, first observed 18 November and last observed 19 November. Its seven day chart prints 0, 0, 0, 0, 0, 6, 3 above the dates 13 to 19 November, which this site reads as zero on 13 to 17 November, 6 on 18 and 3 on 19 (derived: 6 + 3 = 9). The page says its sensor observations "do not show that the target was vulnerable or that the attempt succeeded" and that its "Confirmed" confidence "does not mean the attempt succeeded". It lists a possible proof of concept on a code hosting site, created at 01:57:30 UTC on 19 November, which it calls unverified and says could contain malware, and no scanner coverage. Its request targets, user agents and the six addresses sit behind an Enterprise tier, so none is printed here.
What SecurityWeek relays. SecurityWeek's page data gives 2026-11-19T09:10:58 UTC; the page shows 4:10 AM ET, which is the same moment because US clocks went back on Sunday 1 November (derived: 04:10:58 at UTC-5). This site could not check any of the following at its source, and attributes each to SecurityWeek:
- GreyNoise, as SecurityWeek reads it: as of early 19 November, over 150 requests matching the shape of a Java4Shell attack from five IP addresses, and no successful intrusions on its sensors.
- WatchTowr: its honeypot network saw sprays with User-Agent values containing "java4shell-recon" and blind attempts with correctly shaped requests. Its head of threat intelligence is quoted as saying what it sees "isn't particularly functional or scary yet", and as expecting product by product exploits over the coming days.
- Cloudflare's Cloudforce One: scanners are starting to hide their tools. In one case an address probed with a User-Agent that referenced Java4Shell and returned roughly 17 hours later with the trigger while impersonating a Chrome browser. On 18 November one address was seen for the first time in both Java4Shell activity and Artex activity; SecurityWeek describes Artex as an open source AI penetration testing framework that threat actors have abused.
- VulnCheck and Immersive: VulnCheck has made a full RCE exploit available to its customers, and a researcher at Immersive is reported to have developed a working exploit. SecurityWeek adds that the proof of concept Calif released only triggers the memory corruption.
- Cisco: aware of PoC code, no knowledge of malicious use. This site read Cisco's own advisory and the wording matches (see the vendor section).
Difference (a), reported and not resolved. GreyNoise's page: 187 sessions from five sources since 17 November 00:00 UTC. Previdian's page: 9 attempts from six addresses in four countries since 18 November. SecurityWeek's reading of GreyNoise: over 150 requests from five addresses as of early 19 November. The unit differs (session, attempt, request), the window differs, and no page says how many of GreyNoise's five addresses are among Previdian's six. Per source the figures are 37.4 sessions (derived: 187 / 5) and 1.5 attempts (derived: 9 / 6), which shows the units differ and nothing more. Neither page says an attempt succeeded: GreyNoise prints zero confirmed at its own sensors and Previdian says its data cannot show success.
The windows do not line up: GreyNoise's opens 21 h 32 min before the CVE record
GreyNoise counts from 00:00 UTC on 17 November. The CVE record saved in brief 733 shows publication at 21:32:03 UTC that day, so the window opens 21 h 32 min 03 s before the record was published (derived: 21:32:03 less 00:00:00). The finder's page, as read in brief 733, puts the OpenJDK advisory and Oracle's patch at 1:00pm PT, which brief 733 converted to 21:00 UTC on the assumption of Pacific Standard Time; on that reading the window opens 21 h 00 min before the fixes (derived). To the time this site saved the page, about 12:15 UTC on 19 November, the window is 60 h 15 min long (derived). GreyNoise's page does not say when inside that window each of the 187 sessions fell, and this site's copy has the per source details removed, so it cannot say how many, if any, came before the release. Previdian's chart shows zero for 17 November.
Three timing details are worth keeping in view, and none is resolved here. First, Azul says its fix shipped "in the morning of November 17th", within an hour of the time the embargo lifted, and names no time zone; the finder's page, as read in brief 733, says 1:00pm PT. This site cannot reconcile the two. Second, Azul also says the CVE ID and severity became public before the coordinated release; Previdian's timeline lists its watchlist entry at 11:06 UTC on 17 November, 10 h 26 min 03 s before the CVE record's 21:32:03 UTC publication (derived, and the 11:06 has no seconds). Third, Previdian prints two different times for its first sensor stamp: 01:54 UTC on 18 November in its timeline and 14:37 UTC under "Previdian Sensors First" in its evidence table, 12 h 43 min apart (derived); the page does not say which is the first attempt.
Vendors: who has a fix, who has a workaround, who has neither
Eight vendor pages, each with the time stamp it prints. Cisco and Siemens list no fix. Siemens prints a mitigation; Cisco prints none. Elastic, Red Hat and Atlassian describe fixes that are expected, and Jenkins, Keycloak and Azul point to builds that exist.
Cisco. The advisory is rated "High", Version 1.1, Interim, first published 18 November 16:00 GMT and last updated 18 November 21:16 GMT, about 14 h 59 min before this site read it (derived). Catalyst Center is listed as affected with "Release No. TBD". IOS XE Software and UCS Central Software are "Under investigation". Of the 14 products in the priority table, 1 is affected, 2 are under investigation and 11 are "Not affected" (derived count). A second table lists 12 more affected products, each with a release still marked TBD: Smart PHY, Connected Mobile Experiences, Secure Network Analytics, Secure Workload, Cisco IQ Virtual Appliance, License (Smart Software Manager) On-Prem, cnBNG, Optical Network Planner (releases 26.4.1 and later only), Services as Code, Splunk AppDynamics On-Premises Virtual Appliance, Ultra Cloud Core and Intersight Virtual Appliance. That is 13 affected products in all (derived: 1 + 12), which agrees with SecurityWeek's "more than a dozen". Cisco says "No workarounds available", says x86-64 systems with AVX-512 VBMI are affected and ARM or Cisco Silicon One systems are not, and says its PSIRT is aware that proof of concept code is available and not aware of malicious use.
Siemens. SSA-272816 has a publication date and Last Update of 2026-11-18, version V1.0, and a CVSS v3.1 base score line of 8.1. It lists three products as affected, all versions, with "Currently no fix is available": Industrial Edge Management Pro V1, Pro V2 and Virtual. Industrial Edge Device Kit and Industrial Edge Management Cloud are listed as not affected, because the component is not present. Siemens says it is assessing the impact and preparing fixes, and prints one mitigation, "Block direct internet access to IEM Pro / IEM Virtual", adding that the most effective immediate measure is to shut down instances or disable internet connectivity until a fix is available and applied. That is the same advice SecurityWeek summarises as taking the instances offline.
Elastic, by deployment (difference e). The post is dated 18 November, and its page data shows it modified at 11:11:38 UTC on 19 November. Self-managed Elasticsearch 7.17.3 and later is affected; Elastic expects to ship a bundled JDK 27.0.1.1+4 in 8.19.26, 9.4.10, 9.5.7 and 9.6.0, and prints a mitigation (see the technical section). Logstash 8.6.2 and later is affected only through the Elastic Integration filter when an Elasticsearch ingest pipeline decodes Base64 from an untrusted source, and Elastic says the result is a denial of service, not code execution. Elastic Cloud Serverless was "only vulnerable in extremely rare cases" and was hotfixed during Wednesday 18 November. For Elastic Cloud Hosted, Elastic says approximately under 5 percent of deployments use a vulnerable CPU architecture in Microsoft Azure, that deployments on GCP and AWS are "not affected at this time", and that it is deploying a configuration change. Elastic Cloud Enterprise needs 4.2.2 or 3.8.7 "once released". Kibana, Beats, Elastic Agent, APM Server, the Cloud on Kubernetes operator and Maps Server do not use Java.
Red Hat. The bulletin rates the issue Critical, lists Red Hat Enterprise Linux 8, 9 and 10 and Red Hat builds of OpenJDK 17, 21 and 25 as affected, and says OpenJDK 8 and 11 are not. It says Red Hat is expediting fixes and recommends updating as soon as errata are available. The saved page prints no date or update stamp. Keycloak's page says Oracle and Red Hat have released updates; Red Hat's own bulletin, as read, does not say that. This site does not resolve the two.
Jenkins and Keycloak. The Jenkins post of 18 November says the issue is fixed in the November 2026 Java releases 17.0.21.1, 21.0.13.1 and 25.0.5.1; the fix is in Java, not in a Jenkins release. The Keycloak post was published 18 November and prints "Last updated: 19 November 2026, 08:00 UTC", 1 h 10 min 58 s before SecurityWeek's article (derived). It says the 26.8 container images, tags latest, 26.8, 26.8.2 and 26.8.2-3, now carry Red Hat OpenJDK 21.0.13.1, and that as checked on 18 November the 26.8.2-1 and 26.8.2-2 images bundled an affected build.
Atlassian. The community post says Atlassian is remediating its Cloud products and that no Cloud customer action is required. For Data Center it says OpenJDK classified the flaw as CVSS 8.1 and that Atlassian classified it as CVSS 9.2 for its own use; out-of-cycle patches follow "once third-party Temurin releases are made available". Until then it prints a mitigation for the latest versions of six Data Center products in standard installations: Bamboo, Bitbucket, Confluence, Crowd, Jira and Jira Service Management. The post shows no date; its page data gives creation at 02:24:06 UTC on 18 November and an update at 00:14:18 UTC on 19 November. In short, the page publishes a mitigation and a promise of patches, not fixed builds.
Azul. Azul says it shipped its fix in its November 2026 Critical Security Patch Update within an hour of the coordinated release, that no exploit was made public before the fix, and that Java4Shell is "harder to weaponize quickly than Log4Shell", which "gives defenders some time". The post was published at 11:57:07 UTC on 18 November (page data) and its no-exploitation line is dated "As of the evening of November 17th", so it is older than today's pages. SecurityWeek adds that Amazon and Microsoft have also released patched builds; this site did not read their release notes.
The conditions each page lists, and the count nobody prints
Red Hat's bulletin is the page that lists the conditions as a numbered set that must all be present. The figure draws them as an AND chain. The pages do not agree on the set, and none counts the JVMs that meet it.
Which saved page prints which condition, and where they differ
| Condition | Pages that print it | Where the pages differ |
|---|---|---|
| OpenJDK 17.0.6 or later | Red Hat, Cisco, Keycloak ("from 17.0.6 onward"), Azul Core (17.0.6 and later) | Jenkins says versions "after 17.0.6" and before the fixed build. Siemens says only "an affected JVM". Oracle's matrix, as copied by Previdian and Siemens, lists build numbers including a Java 8 build |
| x86-64 with AVX-512 VBMI | Red Hat, Cisco, Siemens, Keycloak, Jenkins, Elastic (as AVX512_VBMI) | Atlassian's post names no chip condition; its workaround is the flag Red Hat says turns AVX-512 off entirely. Red Hat and Elastic both start at Intel Ice Lake and AMD Zen 4 |
| No lowered AVX level in the JVM flags | Red Hat, Keycloak | Not on the Cisco, Siemens, Elastic or Jenkins pages |
| Attacker controlled Base64 reaches the decoder | Red Hat, Siemens | Elastic ties it to specific Elasticsearch and Logstash paths; Keycloak says its pre-authentication requests decode it |
| A JIT compiled decoder | Siemens only | Not on any other page read here |
Where the pages differ
Each difference is reported and none is resolved; this site does not guess which page is right. Differences (a) on counts and (e) on Elastic's deployments are in the sections above.
(b) Severity. Brief 733 read Oracle's 8.1 under CVSS 3.1 and the 9.2 of Talos and the finder under CVSS 4.0. Previdian prints both numbers: its page headlines "CVSS 9.2 · Critical" and then lists both CVSS V4.0 9.2 Critical and CVSS V3.1 8.1 High with their vectors, copying Oracle's text as well. Azul calls them "the same assessment expressed in two different standards". Atlassian is a different case: it says OpenJDK classified the flaw 8.1 and that Atlassian's own Data Center classification is 9.2. The gap between 9.2 and 8.1 is 1.1 points (derived).
Severity as each saved page prints it, and who assigned it according to the page
| Page | Score or label as printed | Assigned by, as the page says |
|---|---|---|
| Oracle risk matrix and CVE record (brief 733) | CVSS 3.1 base 8.1, attack complexity High | Oracle, the CVE assigner |
| Cisco Talos and the finder (brief 733) | CVSS 4.0 9.2 | Each by its own page |
| Previdian | CVSS 9.2 Critical (V4.0); also V3.1 8.1 High, both with vectors | Not said |
| Azul | 8.1 (High) under CVSS 3.1; 9.2 (Critical) under CVSS 4.0 | Azul's reading: 3.1 is the standard OpenJDK uses; 4.0 is the one Calif and most articles cite |
| Atlassian | OpenJDK: CVSS 8.1. Atlassian, Data Center: CVSS 9.2 | OpenJDK, and Atlassian for its own products; version of the second not stated |
| Cisco | "High" advisory label; no number on the saved page | Cisco |
| Siemens | CVSS v3.1 base score 8.1, vector printed, no severity word | Not said |
| Red Hat | "severity impact of Critical"; no score | Red Hat |
| Elastic, Jenkins, Keycloak | No score printed | Not applicable |
(c) Versions and platforms. Oracle's matrix, as Previdian and Siemens copy it, lists Oracle 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. Red Hat says OpenJDK 8 and OpenJDK 11 "are also not affected" because they lack the vulnerable code path, while Azul lists Azul Prime releases for Java 8, 11, 17, 21 and 25 as affected until fixed. This site does not know whether those are the same code; no page says. Jenkins and Keycloak print fixes for Java 17, 21 and 25 only, while Oracle's matrix lists a 27.0.1.0.3 build and Elastic says it will bundle JDK 27.0.1.1+4. On platforms, Cisco, Red Hat, Jenkins and Elastic all say ARM systems are not affected or not known to be affected, Jenkins adds s390x and ppc64le, and Cisco adds Cisco Silicon One. Red Hat names AWS, GCP and Azure instance families "most likely to be affected"; Elastic says its own Cloud Hosted deployments on GCP and AWS are not affected at this time. Those are different fleets, so they are not a contradiction. Elastic does not print the 17.0.6 floor; it prints product versions (Elasticsearch 7.17.3 and Logstash 8.6.2 and later), which are not JDK versions.
(d) How exploitable. Azul says Java4Shell is harder to weaponise quickly than Log4Shell and that no exploit was public before the fix. The finder's page, as read in brief 733, says there is no universal exploit. SecurityWeek relays WatchTowr's "isn't particularly functional or scary yet", Calif's proof of concept that only triggers the memory corruption, VulnCheck's full RCE exploit for customers and a researcher's working exploit. Cisco says a proof of concept exists. Previdian lists an unverified possible proof of concept created at 01:57:30 UTC on 19 November. Taken together, the relayed reports say a working exploit exists for some customers and researchers and that the traffic seen so far looks like probing; no page this site read at the source says which of the 187 sessions or 9 attempts carried one.
(f) Authentication. Three pages describe it differently and this site does not merge them. Elastic says Remote Code Execution on Elasticsearch needs an authenticated user with at least read access, while a denial of service needs no authentication, and that Logstash can only be crashed. Keycloak says it decodes Base64 on requests that arrive before authentication, so a login requirement does not protect an affected instance. Red Hat says HTTP Basic credentials, SAML artifacts, CSRF tokens and MIME encoded words are "routinely decoded before authentication completes" and that pre-authentication remote code execution has been demonstrated against widely deployed Java applications. GreyNoise's trigger shape sits in an Authorization header, which is the kind of value Red Hat and Keycloak describe being decoded early.
What the pages establish, what they do not, and what changed since brief 733
What the saved pages state, and what none of them states
| Question | Stated, and by whom | Not stated |
|---|---|---|
| Did any attempt succeed? | GreyNoise: Successful Exploitation (0) on its gridpoints. SecurityWeek: GreyNoise has seen no successful intrusions on its sensors. Previdian: its data does not show an attempt succeeded. Cisco: not aware of malicious use, as of 18 November 21:16 GMT | Anything about servers outside those sensors, any victim, any indicator of successful compromise |
| Who is behind the traffic? | Previdian: attacker addresses in CN, DE, RO and US. SecurityWeek relays that Cloudflare saw one address tied to both Java4Shell and Artex activity | Any actor, group or attribution. GreyNoise says its clusters are leads, not attribution |
| Working exploit or reconnaissance? | WatchTowr, via SecurityWeek: reconnaissance User-Agents and a sample that is not yet functional. VulnCheck and Immersive, via SecurityWeek: a working exploit exists. Calif's proof of concept only triggers the corruption | Which sessions or attempts carried a working exploit; any statement at the source of WatchTowr, Cloudflare, VulnCheck or Immersive, none of which this site read |
| How many JVMs are exposed? | The conditions (Red Hat, Cisco, Siemens). Elastic: under 5 percent of its Cloud Hosted deployments on one cloud | Any count of exposed JVMs on AVX-512 VBMI hardware that decode attacker controlled Base64 |
| Is it on CISA's catalogue? | Previdian, read 12:16:57 UTC on 19 November: Not in CISA KEV. Azul, as of the evening of 17 November: not in KEV | Any later change. This site did not read the catalogue file itself |
| Which other Cisco products are affected? | IOS XE Software and UCS Central Software are under investigation | The outcome of either |
| When do fixes arrive? | Elastic: 8.19.26, 9.4.10, 9.5.7 and 9.6.0 are upcoming. Atlassian: after Temurin releases. Siemens: preparing. Red Hat: expediting. Cisco: releases marked TBD | A date for any of them |
What brief 733 read on 18 November and what the pages read on 19 November print
| Question | Brief 733 (read 18 November, about 07:10 to 07:40 UTC) | Pages read on 19 November |
|---|---|---|
| Exploitation | Talos: none observed yet. CISA-ADP entry: exploitation none at 21:53:20 UTC on 17 November | GreyNoise 187 sessions and Previdian 9 attempts, none confirmed successful; SecurityWeek relays a working exploit for customers |
| Vendor statements | None read | Eight vendor pages |
| Fixed builds | No saved page printed a fixed version number | Jenkins and Keycloak: 17.0.21.1, 21.0.13.1, 25.0.5.1. Elastic: JDK 27.0.1.1+4 to be bundled |
| Chip condition | Oracle's row and the CVE record printed none; Talos and the finder printed AVX-512 VBMI | Cisco, Red Hat, Siemens, Keycloak, Jenkins and Elastic print it; Atlassian does not |
| Score | 8.1 (Oracle, CVSS 3.1) against 9.2 (Talos and the finder, CVSS 4.0) | Previdian prints both; Azul calls it one assessment in two standards; Atlassian prints its own 9.2; Cisco says High; Red Hat says Critical |
The checks, flags and fixed builds the vendors print
This section copies the defensive items the saved pages print, each attributed in its caption, and nothing else. It prints no input string, no payload and no exploitation step, and the saved pages print no indicators of successful compromise.
Checking a host. Red Hat's bulletin gives two checks: the Java version, and a CPU flag listing, where the output avx512vbmi means the CPU supports the vulnerable instruction set and no output means it does not. As printed, the pattern in the second command has a backslash before its opening parenthesis, which reads like a typo; this site did not run it. Elastic and Jenkins print a simpler check for the one flag.
java -version
grep -E -o '\(avx512f|avx512bw|avx512vbmi)\>' /proc/cpuinfo | sort -u
grep -c avx512vbmi /proc/cpuinfo
grep -o -w -m1 'avx512vbmi' /proc/cpuinfo
The two interim flags. Red Hat says to start affected JVMs with one of them. The first removes the vulnerable code path while leaving other AVX-512 optimisations in place and is Red Hat's preferred option. The second turns AVX-512 off entirely and is for x86-64 only: Red Hat says JVMs on ARM64 do not recognise it and will fail to start. Elastic says customers must not set the opposite form, -XX:+UseBASE64Intrinsics, because it re-enables the optimisation. Atlassian pairs the second flag with an option that ignores unrecognised VM options; its post does not say why, and this site's reading, not stated by the page, is that it protects against the start failure Red Hat describes.
-XX:-UseBASE64Intrinsics
-XX:UseAVX=2
-XX:UseAVX=2 -XX:+IgnoreUnrecognizedVMOptions
Atlassian's post then gives per product steps (startup scripts, environment variables, Helm values, Bitbucket's Mesh and bundled search server, Confluence's Synchrony and conversion sandbox). Those are vendor scripts and settings and are described here, not copied; take them from the post.
Confirming the flag is live. Keycloak and Red Hat both say the environment variable is not enough and the running JVM must report the flag. Keycloak prints the command with a process id placeholder; Red Hat prints the expected output.
jcmd <keycloak-pid> VM.flags -all | grep -F UseBASE64Intrinsics
bool UseBASE64Intrinsics = false
Elastic's self-managed mitigation. Elastic says to create an options file in the jvm.options.d directory on each affected node containing the single flag line, restart the node (a rolling restart avoids downtime), and confirm the flag appears in the node's JVM input arguments through the nodes API. Its second option is to point Elasticsearch at the latest JDK, which Elastic says does not work with Docker.
config/jvm.options.d/cve-2026-114938.options
-XX:-UseBASE64Intrinsics
GET _nodes/jvm
Fixed builds and image tags, as the pages print them. The lines below are laid out by this site; every value is as printed.
Java builds with the fix (Jenkins, Keycloak): 17.0.21.1, 21.0.13.1, 25.0.5.1
Keycloak 26.8 image tags with the fixed runtime: latest, 26.8, 26.8.2, 26.8.2-3
Keycloak image tags that bundled an affected OpenJDK 21.0.13 on 18 November: 26.8.2-1, 26.8.2-2
Elastic bundled JDK it expects to ship: 27.0.1.1+4 in 8.19.26, 9.4.10, 9.5.7, 9.6.0
Azul Prime releases with the fix: 26.08.110.0, 26.02.510.0
Cloud instance families. Red Hat says the families most likely to be affected include the following. It is a pointer for where to check the CPU flag, not a list of vulnerable hosts.
AWS C6i/M6i/R6i and C7i/M7i/R7a
GCP C3/C3D/C4/C4D/N4/N4D/M3/M4
Azure Dsv5/Dsv6/Dasv6/Dasv7 and related D-series v5 to v7 SKUs
What to look for in logs and at the edge. Two descriptions are printed, one by GreyNoise as its definition and one relayed by SecurityWeek from WatchTowr. Cisco's advisory also lists Snort rule ids under its action links. None of them is an indicator of success.
GreyNoise: HTTP Authorization header carried a Basic or Bearer value of the Java4Shell trigger shape (64 or 128 characters plus 1 to 3 more, ending in "=")
WatchTowr, via SecurityWeek: User-Agent values containing java4shell-recon
Snort Rule 1:67502-1:67513
Snort Rule 1:301754
What to do, in order
Take this with you
Numbered checklist; the first four rest on the pages, the rest are this site's suggestions
- Red Hat says: inventory which Java runtime, vendor and full build version, each application uses, including runtimes bundled in containers, product installations and vendor appliances. For a product with its own JVM, use the vendor's supported path rather than patching the JVM alone.
- Red Hat and Cisco say: check x86-64 hosts for AVX-512 VBMI with a printed command; ARM64 and older x86-64 are not known to be affected. Record the result per host, including the cloud instance families Red Hat names.
- Jenkins and Keycloak say: update to a fixed runtime (17.0.21.1, 21.0.13.1 or 25.0.5.1, or your vendor's build that carries it) and restart. Keycloak adds that a product version number alone does not show whether a deployment is fixed.
- Red Hat says: until then use -XX:-UseBASE64Intrinsics, test it first, and confirm it on the live JVM. Use -XX:UseAVX=2 on x86-64 only, never on ARM64.
- This site's suggestion: for products with no fix, follow the vendor page and re-read it, because they change. Cisco prints no workaround; Siemens says to block internet access or shut down; Elastic and Atlassian print flags.
- Red Hat says to prioritise internet facing applications and authentication services; this site adds identity providers such as Keycloak, whose own page says a login does not protect an affected instance.
- This site's suggestion: review logs since 00:00 UTC on 17 November for malformed Authorization headers and for User-Agent values containing java4shell-recon, and treat "attempts" as scanning until a source says otherwise.
- This site's suggestion: do not read GreyNoise's zero as safe. It is one sensor network's view and says nothing about your servers.
What this site worked out
Every derived figure in this brief, with its arithmetic
| Figure | Arithmetic | Note |
|---|---|---|
| 187 sessions | 70 + 64 + 25 + 22 + 6 = 187 | GreyNoise prints five counts and no total |
| 37.4 and 1.5 per source | 187 / 5 = 37.4; 9 / 6 = 1.5 | Different units; shown only to say they differ |
| Previdian daily total | 0 + 0 + 0 + 0 + 0 + 6 + 3 = 9 | Text order read as 13 to 19 November |
| GreyNoise window to the save | 17 Nov 00:00 to 19 Nov about 12:15 UTC = 60 h 15 min | Save time is about 12:15 UTC |
| Window opens before the CVE record | 21:32:03 less 00:00:00 = 21 h 32 min 03 s | Record time from brief 733's saved copy |
| Window opens before the fixes | 21:00 less 00:00 = 21 h 00 min | 1:00pm PT read as PST, as brief 733 did |
| SecurityWeek in UTC | 04:10:58 at UTC-5 = 09:10:58 UTC | Page data agrees; clocks went back 1 Nov |
| Cisco affected products | 1 + 12 = 13; priority table 1 + 2 + 11 = 14 | Counts of the printed rows |
| Cisco advisory age | 19 Nov 12:15 less 18 Nov 21:16 = 14 h 59 min | Last Updated stamp |
| Cisco update span | 21:16 less 16:00 = 5 h 16 min | First published to last updated |
| Previdian stamp gap | 14:37 less 01:54 = 12 h 43 min | Two stamps for the first sensor observation |
| Previdian watchlist before the record | 21:32:03 less 11:06 = 10 h 26 min 03 s | The 11:06 has no seconds |
| Keycloak update before SecurityWeek | 09:10:58 less 08:00:00 = 1 h 10 min 58 s | Both stamps from page data or page text |
| Score gap | 9.2 less 8.1 = 1.1 | Two CVSS versions |
The question the pages leave open
Every page agrees on the narrow condition, an x86-64 chip with AVX-512 VBMI running an affected OpenJDK, and every counter says its own number is not a success. None says how many JVMs meet the condition, and the only figure for success is zero on one network's sensors. If that is the only count of success, what is yours, and who is looking for it in your logs?
Key facts
Sources
- PrimaryGreyNoise Java4Shell tracker: the five source count, the Authorization header definition, sessions per source, Successful Exploitation (0). Address list removed by this site. Read at about 12:15 UTC on 19 November 2026; robots.txt answered 403GreyNoiseaccessed 2026-11-19
- PrimaryPrevidian vulnerability report for CVE-2026-114938: 9 attempts, 6 IPs, 4 countries, 5 sensors, seven day chart, CVSS 9.2 and 8.1, evidence table and timeline. A vendor product page. Rendered copy saved 12:16:57 UTC on 19 November 2026Previdianaccessed 2026-11-19
- PrimaryCisco Security Advisory cisco-sa-nov2026-java-rce-eCa6ss9z: High, Version 1.1 Interim, last updated 2026 November 18 21:16 GMT, affected and under investigation tables, no workarounds, Snort rules. Read on 19 November 2026Ciscoaccessed 2026-11-19
- PrimarySiemens SSA-272816: three Industrial Edge Management products, no fix, mitigation, CVSS v3.1 8.1, Last Update 2026-11-18. Read on 19 November 2026Siemens ProductCERTaccessed 2026-11-19
- PrimaryElastic blog post of 18 November 2026: Elasticsearch, Logstash, Cloud Serverless, Cloud Hosted, ECE and ECK, the self-managed mitigation. Read on 19 November 2026; page data modified 11:11:38 UTCElasticaccessed 2026-11-19
- PrimaryRed Hat bulletin RHSB-2026-018: Critical, affected products, the four conditions, the CPU check, the two interim flags, cloud instance families. The page prints no date. Read on 19 November 2026Red Hataccessed 2026-11-19
- PrimaryAzul blog post of 18 November 2026: fix shipped 17 November, fixed Azul Core and Prime builds, CVSS 8.1 and 9.2, harder to weaponise than Log4Shell. A Java vendor; its own survey is not repeated. Read on 19 November 2026Azulaccessed 2026-11-19
- PrimaryJenkins blog post of 18 November 2026: fixed Java builds 17.0.21.1, 21.0.13.1 and 25.0.5.1, the flag and the CPU check. Read on 19 November 2026Jenkins projectaccessed 2026-11-19
- PrimaryKeycloak post of 18 November 2026, last updated 19 November 2026 08:00 UTC: updated image tags, affected tags, checks and the flag. Read on 19 November 2026Keycloakaccessed 2026-11-19
- PrimaryAtlassian Community post on Data Center mitigations: Cloud statement, CVSS 9.2 for Data Center, patches pending, the UseAVX flag mitigation. Read on 19 November 2026Atlassianaccessed 2026-11-19
- PrimaryOracle Critical Security Patch Update Advisory, November 2026, CVE-2026-114938 row: CVSS 3.1 8.1 and the affected build numbers. Read in brief 733 on 18 November 2026Oracleaccessed 2026-11-18
- PrimaryCisco Talos threat advisory on CVE-2026-114938: CVSS 4.0 9.2, no exploitation observed yet as of its time. Read in brief 733 on 18 November 2026Cisco Talosaccessed 2026-11-18
- PrimaryThe finder's write-up: CVSS 4.0 9.2, no universal exploit, the 1:00pm PT release time. Read in brief 733 on 18 November 2026Calif Research (Calif.io)accessed 2026-11-18
- PrimaryCVE-2026-114938 record: published 2026-11-17T21:32:03Z, CVSS 3.1 8.1. Read in brief 733 on 18 November 2026CVE Program (assigned by Oracle)accessed 2026-11-18
- Reported bySecurityWeek, 19 November 2026 4:10 AM ET (page data 09:10:58 UTC): relays WatchTowr, Cloudflare, VulnCheck, Immersive and Cisco statements. A secondary report used only for what it relays, each item attributed. Read on 19 November 2026SecurityWeekaccessed 2026-11-19


