What happened
On 17 November Calif disclosed Java4Shell, CVE-2026-114938: memory corruption in the Base64 decoder of OpenJDK’s HotSpot JVM (vendor, Calif). The same day Oracle shipped its November Critical Patch Update and the OpenJDK Vulnerability Group published an advisory covering the 17, 21, 25, and 27 release lines (independent, Tenable).
The scores disagree, and the gap tells you something. Oracle rates it 8.1 under CVSS 3.1 with high attack complexity (independent, Tenable). AWS rates it 8.8 (vendor, AWS). Calif gives 9.2 under CVSS 4.0 (vendor, Calif). As of 17 November there were no reports of exploitation and it was not in CISA’s KEV catalogue; Immersive’s Kevin Breen says his team built a working exploit and has not released it (independent, Tenable).
Three conditions, not one
Coverage will call this the next Log4Shell. The exposure test is narrower. A JVM is in scope only when all three of these hold, which is the point of Fig 1.
| Condition | What is reported | Label |
|---|---|---|
| An affected JDK build | OpenJDK 17.0.6 and later; the OpenJDK advisory names the 17, 21, 25, and 27 lines. Amazon Corretto 8 and 11 are listed as not affected. | vendor, independent, vendor |
| A CPU with the right instructions | x86-64 reporting avx512f, avx512bw, avx512vbmi, and bmi2: Intel Xeon from Ice Lake on, AMD Zen 4 or newer. ARM64 is not affected. | vendor, vendor |
| A path that decodes attacker Base64 | Calif reports code execution before login in its tests on Keycloak, Jira, Confluence, Artifactory, Jenkins, and PingFederate, and after login on Kafka, Solr, Elasticsearch, Gerrit, and Spinnaker. It says the list is not exhaustive and the product chains are withheld. | vendor, via Tenable |
The mechanism is specific. For a 65-byte input the fast decoder allocates a 50-byte array and writes 64 bytes from offset 48, so the input is rejected only after neighbouring heap memory has been overwritten (vendor, Calif). Calif also says there is no universal exploit and that the heap grooming is timing-sensitive. Treat that as a reason to rank, not a reason to wait.
Fixed builds, and the October gap
Oracle’s fixed builds are 8u513-perf, 17.0.21.1, 21.0.13.1, 25.0.5.1, and 27.0.1.1 (independent, Tenable). On 9 November Ubuntu moved OpenJDK 17 to 17.0.21 (independent, LWN, USN-9014-1). That build sits inside the affected range and below the 17.0.21.1 fix, so a server patched in the October cycle can still be exposed (our maths). On 17 November AWS listed Amazon Corretto 17, 21, 25, and 27 as pending a fix (vendor, AWS). For Temurin, Zulu, Microsoft, Red Hat, and the rest, read each vendor’s advisory; we have not.
Patching is the vendor’s job. Knowing which JVMs you actually run is yours.
Stopgap flags
Calif lists two JVM startup flags (vendor, Calif):
-XX:-UseBASE64Intrinsicsturns off the fast Base64 path. It works on any architecture, so it is the one to use in a mixed fleet.-XX:UseAVX=2turns off AVX-512, on x86-64 only. A JVM on ARM will not start with it, so a shared launch script that adds it everywhere takes your ARM services down.
Neither Oracle nor the OpenJDK advisory mentions these flags (independent, Tenable). We have not measured what either costs in throughput. Use them where the patched build is days away, and remove them when it lands.
Find exposed JVMs
Check the process that is running, not the java on the path. Endor Labs recommends reading the running binary and asking the live JVM for its version with jcmd (independent, Endor Labs). The script below combines that with the CPU test. Unexecuted script: Prestanda has not run it on a fleet. Read it before you run it.
#!/usr/bin/env bash
# Java4Shell triage, per host. Prints CPU exposure, then each running JVM.
flags=$(grep -m1 '^flags' /proc/cpuinfo)
cpu=exposed
for f in avx512f avx512bw avx512vbmi bmi2; do
echo "$flags" | grep -qw "$f" || cpu="not exposed (no $f)"
done
echo "$(hostname) $(uname -m) cpu: $cpu"
for pid in $(pgrep -f java); do
bin=$(readlink -f /proc/$pid/exe)
ver=$(jcmd $pid VM.version 2>/dev/null | grep -m1 -E 'JDK [0-9]')
base64=$(tr '\0' ' ' < /proc/$pid/cmdline | grep -o -- '-XX:-UseBASE64Intrinsics')
echo " pid $pid $bin $ver ${base64:-no flag}"
doneA host whose CPU line says “not exposed” drops to normal patch cadence. Inside a container, /proc/cpuinfo shows the host’s CPU, so the answer follows whichever node the workload lands on. Check the node pool, not the image.
Five steps this week
- Step 1. Inventory running JVMs and CPU flags. Run the check on every x86-64 host and node. Owner: platform engineering.
- Step 2. Rank by exposure. Internet-facing services that decode Base64 before login first: identity providers, CI, artefact stores, and wikis. Owner: application security.
- Step 3. Add
-XX:-UseBASE64Intrinsicswhere the fix is not yet available. Record each host so the flag comes out with the patch. Owner: platform engineering. - Step 4. Patch and rebuild. Move to your vendor’s fixed build, rebuild container base images, and confirm the running build with
jcmd. Owner: platform engineering with each service owner. - Step 5. Watch for exploitation. Track CISA KEV and your JDK vendors’ advisories daily until the fleet is patched. Owner: security operations.
Missing
- The OpenJDK Vulnerability Group advisory, read directly rather than through Tenable.
- Fixed builds for Temurin, Zulu, Microsoft, Red Hat, Ubuntu, and Corretto.
- Calif’s product exploit chains, which it has withheld.
- An independent reproduction. Endor Labs did not reproduce it.
- The throughput cost of either stopgap flag.
Frequently asked questions
Is every Java server exposed to Java4Shell?
No. Calif says three things must line up: an affected OpenJDK build (17.0.6 or later on the 17 line, and the 21, 25, and 27 lines per the OpenJDK advisory), an x86-64 CPU that reports avx512f, avx512bw, avx512vbmi, and bmi2, and a code path that decodes attacker-supplied Base64. ARM64 servers are not affected, according to AWS.
Which Java builds fix Java4Shell?
Tenable lists Oracle's fixed builds as 8u513-perf, 17.0.21.1, 21.0.13.1, 25.0.5.1, and 27.0.1.1. Other distributions publish their own fixes; on 17 November AWS listed Amazon Corretto 17, 21, 25, and 27 as pending a fix. Check your vendor's advisory for the exact build.
What can we do before the patch reaches us?
Calif lists two JVM startup flags. -XX:-UseBASE64Intrinsics turns off the fast Base64 path. -XX:UseAVX=2 turns off AVX-512, but only on x86-64; Calif warns that a JVM on ARM will not start with it. Neither Oracle nor the OpenJDK advisory mentions these flags, and Prestanda has not measured their performance cost.
Living page. Last updated 18 Nov 2026. We update this note when vendors publish fixed builds, exploitation is reported, or the OpenJDK advisory changes. Need a fleet-wide JVM inventory this week? Talk to us.