On November 17, 2026, Oracle and the OpenJDK project disclosed CVE-2026-114938, known as Java4Shell. It is a pre-authentication remote code execution vulnerability in OpenJDK’s hand-optimized Base64 decoding routine. Affected versions include Java 17.0.6 and later, as well as Java 21, 25, and 27, when they run on x86-64 servers that support AVX-512 VBMI instructions.
Because the vulnerability is in the Java runtime, it reaches every application built on an affected version. This includes widely deployed products such as Keycloak, Jira, Confluence, Jenkins, JFrog Artifactory, and PingFederate. Attackers can exploit many of these products before a user logs in.
The patch was released alongside the disclosure, and the exposure is immediate. Opportunistic exploit attempts are already being seen in the wild, though no successful compromise has been confirmed. A single malformed HTTP request can trigger the vulnerability. Proof-of-concept code that crashes the runtime is already public. The researchers who found it also have working exploits for specific applications, which they have not released yet. Upgrade the Java runtime as soon as possible. Until every server is patched, AppTrana can block malicious requests that target exposed applications.
Java4Shell (CVE-2026-114938): Risk Analysis
What Is Java4Shell?
Java4Shell is a heap overflow in the AVX-512 routine that HotSpot substitutes for java.util.Base64 decoding once that code path runs often enough to be JIT-compiled. When the routine handles the final partial block, it subtracts one byte for each trailing = padding character. A malformed input whose length is one to three bytes past a multiple of 64 drives this calculation negative. The store mask is then read as full width, and the routine writes 62 to 63 fixed bytes past the end of the destination array on the heap. Java only rejects the input as invalid after that out-of-bounds write has already happened.
Three conditions must all hold for the vulnerability to be reachable:
- The CPU is x86-64 with AVX-512 VBMI (Intel Ice Lake or newer, AMD Zen 4 or newer). ARM64 is not affected.
- The attacker-controlled data reaches the basic or URL decoder of java.util.Base64. The MIME decoder is not affected, and neither are Jackson or Apache Commons Codec, which ship their own decoders.
- The decode path has been exercised enough to be JIT-compiled, which on a busy server can happen within seconds.
The reason this is a pre-authentication problem is that several common HTTP inputs are Base64-decoded before any credential check:
- HTTP Basic credentials in the Authorization header, decoded before the password is verified.
- JWT segments, decoded by libraries such as Auth0 java-jwt that use java.util.Base64.
- SAML artifacts (SAMLRequest, SAMLResponse, SAMLart) on identity and service provider endpoints.
The attacker can choose where the overflow bytes land in memory. The content of those bytes is outside the attacker’s control. As a result, reliable code execution depends on the target’s heap layout and timing, and failed attempts often crash the JVM. This raises the attack complexity.
Affected and Fixed Versions
The following Oracle builds are confirmed affected, with the fixed build for each line. OpenJDK 17.0.6 and later, 21, 25, and 27 share the same root cause.
| Java line | Affected (example) | Fixed build |
| Java 8 Enterprise Performance Pack | 8u511b33-perf | 8u513-perf |
| Java 17 | 17.0.6 through 17.0.21.0.3 | 17.0.21.1 |
| Java 21 | up to 21.0.13.0.3 | 21.0.13.1 |
| Java 25 | up to 25.0.5.0.3 | 25.0.5.1 |
| Java 27 | up to 27.0.1.0.3 | 27.0.1.1 |
| Oracle GraalVM for JDK 17 / 21 | 17.0.21.0.3 / 21.0.13.0.3 | See Oracle advisory |
The OpenJDK Vulnerability Group confirmed CVE-2026-114938 in its advisory on November 17, 2026. Standard Java 8 and Java 11 are not listed as affected.
Downstream distributions are built from the same OpenJDK source. These include Amazon Corretto, Eclipse Temurin, Azul Zulu, BellSoft Liberica, SAP SapMachine, Microsoft Build of OpenJDK, and Red Hat OpenJDK. Each vendor releases fixed builds on its own schedule, so check the vendor’s advisory for the exact patched version.
The same guidance applies to Java runtimes bundled inside container images and appliance firmware.
Exception: a host is only exploitable when it runs an affected build on an x86-64 CPU with AVX-512 VBMI. ARM64 hosts and older x86-64 hosts without AVX-512 VBMI run the vulnerable code but never reach the affected intrinsic.
What Java4Shell Means for Your Web Infrastructure
- Pre-authentication takeover – Where exploitation succeeds, an attacker runs code on the server with no credentials, leading to full compromise of the affected product and the data it holds.
- Credential and data exposure – A foothold in an identity product such as Keycloak or PingFederate puts tokens, secrets, and federated trust at risk across every application behind it.
- Denial of service – Even without code execution, the overflow crashes the JVM. Repeated triggers against an internet-facing service become an easy, low-effort outage.
- Wide blast radius – The vulnerability is in the runtime, so one unpatched Java version exposes every application on it, including third-party and appliance software that embeds its own JRE.
- Quiet probing – Early exploitation attempts look like malformed authentication traffic and failed logins, so they blend into normal noise unless you are watching for the specific signature.
From Disclosure to In-the-Wild Attempts
| Date | Milestone |
| October 22, 2026 | CVE-2026-114938 reserved |
| November 17, 2026 | Oracle November CPU (Critical Patch Update) and the CVE record published; OpenJDK advisory and the reporter’s technical write-up released the same day; a crash-level PoC goes public |
| November 18, 2026 | Multiple security labs independently reproduce the overflow and JVM crashes; no public weaponized RCE yet, and the reporter withholds the per-application exploits |
| November 19, 2026 | Opportunistic exploit attempts observed in the wild (GreyNoise); CISA issues a patch-now alert; NVD ingests the CVE (Undergoing Analysis); OX Security estimates that under 0.12% of JVMs face genuine RCE risk |
How to Check Your Exposure to Java4Shell
Find every affected Java runtime, including JREs bundled inside applications and containers, and confirm which hosts run on vulnerable CPUs. On Linux, all four flags below must be present for the host to be exploitable:
grep -m1 '^flags' /proc/cpuinfo | tr ' ' '\n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'
Java4Shell: Mitigation and Remediation
Patch first – Upgrade every affected runtime to its fixed build, then restart each JVM, because a running process keeps the old runtime in memory. Rebuild and redeploy container images on an updated base image and ask any vendor that bundles a JRE when their fixed build ships.
Interim JVM flags – Where you cannot patch immediately, two flags keep the JVM off the vulnerable path. Neither is a vendor-confirmed workaround for this CVE, and both carry a performance cost, so test before production use and remove them after patching:
# Disable the vulnerable Base64 intrinsic
-XX:-UseBASE64Intrinsics
# x86-64 only: keep the JVM off the AVX-512 path (ARM JVMs will not start with this)
-XX:UseAVX=2
Verify a running process with jcmd <PID> VM.flags -all.
Reduce exposure – Remove Java services from direct internet exposure where you can, and place admin interfaces behind a VPN or an allow list.
Virtual-patch at the edge – A WAF in block mode can reject the malformed-Base64 trigger before it reaches the runtime, which buys time across a large fleet. A WAF is a compensating control; it cannot see the Java-TLS-client path, Kafka wire protocol, other non-HTTP inputs, or fields it does not decode. Keep patching.
How AppTrana WAAP Protects You
Patching the Java runtime is the permanent fix for Java4Shell. In large environments with many Java applications, that rollout takes time. AppTrana protects internet-facing applications during this period. It blocks attack attempts on exposed HTTP paths from the moment the CVE is understood, and that protection stays in place until every server is patched.
- Block mode from day zero – AppTrana runs in block mode, so malformed and injection-style inputs are rejected out of the box, with no learning-mode window during which the application stays exposed.
- Emergency virtual patch – AppTrana’s research team ships an app-specific rule that blocks the Java4Shell trigger, the malformed Base64 string, across the Authorization header, JWT, and SAML parameters.
- 24×7 managed monitoring – AppTrana’s managed services team watches for the request and crash signatures tied to this CVE and tunes the response as exploitation techniques evolve.