Java4Shell (CVE-2026-114938) is a memory-corruption vulnerability in the hardware-accelerated Base64 decoder of OpenJDK’s HotSpot JVM. A crafted Base64 string ending in padding causes the decoder to write past the end of its output buffer on the heap. Depending on how the application handles that input, an unauthenticated remote attacker can read sensitive heap memory and, in demonstrated cases, execute code on the server.
Oracle scores it 8.1 (High) under CVSS v3.1, reflecting high attack complexity. Calif demonstrated pre-authentication remote code execution on default installs of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate.
Two conditions must both be true for a system to be exposed: it runs an unpatched OpenJDK-based runtime from 17.0.6 onward (from any vendor), and it runs on an x86-64 processor with AVX-512 VBMI. Oracle and the OpenJDK project released fixes on 17 November 2026. There is no confirmed exploitation in the wild yet, but technical details leaked before the patch and trigger code is already public.
Where to start:
patch internet-facing identity providers and developer platforms first. Where a patch has to wait, start the JVM with -XX:-UseBASE64Intrinsics and restart it.
9.2 Critical (CVSS v4.0, Calif)8.1 High (CVSS v3.1, Oracle)
Impact
Heap memory disclosure; remote code execution, pre-auth in some default installs
Affected
OpenJDK-based runtimes 17.0.6+ on x86-64 CPUs with AVX-512 VBMI
Fix released
17 November 2026 (Oracle Critical Patch Update and OpenJDK advisory)
Exploited in the wild
Not confirmed as of 18 November 2026; not in CISA KEV
What’s at Risk
Base64 decoding is everywhere in Java software. Authentication headers, session and API tokens, identity assertions, certificates, file uploads and API payloads all routinely carry Base64-encoded data. That means the vulnerable code path is often reachable with data an attacker controls, sometimes before any login takes place.
How the flaw works
On processors that support AVX-512 VBMI, HotSpot replaces Java’s standard Base64 decoding with a hand-written assembly routine (an “intrinsic”) for speed. When input ends with = padding, a length calculation in that routine underflows. The resulting write mask lets the decoder write a full 64-byte block instead of the intended few bytes, corrupting roughly 62 bytes of adjacent heap memory. Reported triggers are as simple as 64 “A” characters followed by a single “=” (some older Java 17 builds need 128).
Applications shown to be exploitable
Calif tested popular Java applications in their default configurations. The results below show where the flaw has been proven, not the full list of what is exposed.
Access needed
Applications
No authentication (pre-auth RCE)
Keycloak
Atlassian Jira
Atlassian Confluence
JFrog Artifactory
Jenkins
PingFederate
Authenticated user
Apache Kafka Broker
Apache Solr
Elasticsearch
Gerrit
Spinnaker
Sample applications
Apps built on Apache Struts and Spring Boot
The pre-auth list is the most worrying. Identity providers and single sign-on servers hold the keys to everything behind them. Build servers, source repositories and artifact stores sit in the software supply chain, so a compromise there can spread to every system they deploy to.
More than code execution
Remote code execution is not the only outcome. The same bug gives attackers a way to read heap memory, which may expose private keys, session tokens, passwords and API credentials even where full code execution is not achieved. Calif also describes a client-side variant: a Java program that opens a TLS connection to a malicious server can itself be compromised. Outbound connections from Java services are therefore part of the attack surface too.
Are You Affected?
A system is in scope only when both the runtime and the processor conditions below are met, and attacker-controlled data can reach the vulnerable decoding path.
Condition
Affected
Not affected
Java runtime
!Unpatched OpenJDK-based builds from 17.0.6 onward, including the 21, 25 and 27 lines. Oracle Java SE 8 Enterprise Performance Pack 8u511b33-perf.
✓Releases before 17.0.6, including standard Java 8 and 11. Any vendor build that contains the fix.
Distributions
!Oracle Java SE and GraalVM for JDK, Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine, Red Hat OpenJDK, and other OpenJDK-based builds.
✓Builds patched under each vendor’s own advisory.
Processor
!x86-64 with AVX-512 VBMI: Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids, Granite Rapids; AMD EPYC and Ryzen on Zen 4 or newer.
✓ARM64 (including cloud ARM instances). x86-64 without AVX-512 VBMI, including older Intel server chips and most Intel laptops and desktops.
Check the hardware
On Linux, the host meets the processor condition only if all four of these CPU flags are present:
Many commercial products and container images ship their own JDK. Updating the operating system’s Java package does not patch those bundled copies.
Use the CPU condition to prioritise, not to close findings.
A workload on unaffected hardware today can land on an AVX-512 host after a resize, migration or autoscaling event. Treat an unpatched runtime as a finding regardless of where it runs today.
Exploitation Status
As of 18 November 2026, no exploitation in the wild has been confirmed, and CVE-2026-114938 is not in CISA’s Known Exploited Vulnerabilities catalog. No threat actor, targeted sector or region has been publicly linked to the flaw. CISA nonetheless issued an alert on 17 November urging organisations to patch immediately.
That quiet period is unlikely to last. Technical details leaked publicly before the patch was released, and trigger-only code circulated on the day of disclosure. Calif has published a bounded proof of concept that confirms the memory overwrite without taking control of a system. It is withholding its application-specific exploit chains to give defenders time to upgrade. Separately, at least one other researcher has reported building a working, application-specific exploit that has not been released.
The speed of discovery is also a warning. The researcher reportedly found the bug with an AI coding assistant in about an hour and had a first working exploit roughly two hours later, for under $200 in compute. Other teams report building proofs of concept in a similar timeframe. Attackers have access to the same tools.
There is no universal exploit. Each target application needs its own profile for delivering the malicious Base64 data, and exploitation is sensitive to timing and network conditions. That slows attackers down, but it does not stop them, especially against widely deployed applications with predictable default configurations. Do not plan around the absence of exploitation.
Timeline
9 Sep 2026
A related malformed Base64 input appears in a public OpenJDK pull request. No source has tied it to the later leak.
Before 17 Nov
Technical details of the flaw leak publicly ahead of the coordinated release.
17 Nov 2026
OpenJDK advisory and Oracle fixes published (1:00 pm PT). Calif. publishes its write-up and a bounded proof of concept. Trigger-only code circulates publicly. CISA issues an alert urging immediate patching.
18 Nov 2026
No confirmed in-the-wild exploitation; not listed in CISA KEV.
What to Do Now
1
Apply the fixed builds
Oracle released the following fixed Java SE builds on 17 November 2026:
Java SE version
Fixed build
8 Enterprise Performance Pack
8u513-perf
17
17.0.21.1
21
21.0.13.1
25
25.0.5.1
27
27.0.1.1
Other OpenJDK vendors publish fixes on their own schedules; Amazon, for example, released Corretto 17.0.21.1.2 on 17 November. Check each vendor’s advisory for the fixed build number and confirm it addresses CVE-2026-114938 specifically. GraalVM and products that bundle a JDK also need their own updates.
2
If you cannot patch yet, apply a temporary workaround
Preferred:
-XX:-UseBASE64Intrinsics. This turns off the hardware-accelerated Base64 decoder where the bug lives and falls back to the standard Java implementation. Applications that decode large volumes of Base64 may slow down, so test before rolling out widely.
Alternative:
-XX:UseAVX=2. This disables AVX-512 for the whole JVM, so the performance impact is broader. Use it only on x86-64 hosts: JVMs on ARM do not recognise the option and will fail to start.
Neither flag is a fix. Oracle and the OpenJDK advisory do not list them as official mitigations; treat them as stopgaps until the patched build is deployed.
3
Restart and verify
A patch or JVM flag takes effect only after the Java process restarts. Restart every affected service, then confirm the running process uses the fixed build or carries the flag. A correct installed package does not prove the running JVM is safe. Rebuild container images that bake in a JDK, rather than relying on host updates.
4
Reduce exposure while you work
Restrict external access to unpatched Java services where the business allows it, especially administrative and login endpoints.
Do not rely on a reverse proxy or WAF as a control. Request buffering and filtering can break some exploit chains, but the vulnerable decoder can be reached in too many ways for that to be a general mitigation.
Watch for unexpected JVM crashes and fatal error logs on exposed services. Failed exploitation attempts against a memory-corruption bug can show up as crashes before anything else.
Consider rotating keys, tokens and credentials held by internet-facing Java services that ran unpatched, given the heap-disclosure risk.
5
Work in priority order
1Internet-facing identity and SSO services, such as Keycloak and PingFederate.
2Internet-facing developer and collaboration platforms: Jira, Confluence, Jenkins and Artifactory.
3Other internet-facing Java services that handle logins, tokens, certificates or file uploads.
4Internal Java systems, starting with Kafka, Solr, Elasticsearch, Gerrit and Spinnaker.
5Everything else running an affected runtime, including client applications that make outbound TLS connections.
Give every system that is waiting on a vendor build a named owner, an interim mitigation and a remediation deadline, and track it to closure.
How Hive Pro customers are covered
Hive Pro Exposure Management customers have had visibility into Java4Shell (CVE-2026-114938) since day one of the disclosure, through three capabilities of the platform:
Scanless discovery:
Identify vulnerable Java instances directly from software inventory, without scanning a single system.
Vulnerability scanning:
Hive Pro’s scanners actively detect vulnerable Java instances across the environment.
BAS Attack:
Validate exploitability with safe, controlled attack simulations and focus remediation where it matters most.
HiveForce Labs advisory:
HiveForce Labs, Hive Pro’s threat intelligence unit, has published a Java4Shell advisory on the platform that customers can use to pinpoint affected assets.
What Does Rapid7 Do? An Enterprise Checklist for Security Teams
Learn what Rapid7 does across cybersecurity, vulnerability management, and security operations, then use this enterprise checklist to assess fit and workflows.