1. Basic Information
- Article Name: Java4Shell: Write once, RCE everywhere
- Publisher: Calif
- Publication Date: November 17, 2026
- Source: Calif
- Related Sources: OpenJDK Official Advisory, Oracle Official Advisory, IPA, CERT-EU, VulnCheck, Red Hat
- Related CVE: CVE-2026-114938
- Affected Products: OpenJDK / Oracle Java SE 17, 21, 25, 27, Java SE 8 performance builds, and others
- Severity: Critical (Author Priority)
- Official Rating: Ratings differ by source. Calif, the discoverer, scores it 9.2 (Critical) under CVSS 4.0; Red Hat assigns its own CVSS v3 score of 9.8; Oracle's CVSS v3.1 base score is 8.1.
2. Executive Summary
On x86-64 environments featuring AVX-512 VBMI, specially crafted Base64 input reaching the vulnerable decoder intrinsic can cause memory corruption, leading to a denial of service (DoS) or, under specific conditions, pre-authentication RCE.
3. Attack Flow
The following are independent product-specific paths where Calif demonstrated pre-authentication RCE in a controlled environment. In all paths, input must reach the Base64 intrinsic on an affected JVM and supported CPU. RCE additionally requires product-specific exploitation steps, such as combining heap layout manipulation with information disclosure. Product-specific exploit chains have not been publicly released; the following is a summary of the published common mechanism.
Path A: Atlassian Jira
- Attacker input reaches the affected intrinsic via Jira's Base64 processing.
- The demonstration combines memory corruption with product-specific techniques such as heap manipulation and information disclosure to achieve code execution with the privileges of the JVM process.
Path B: Jenkins
- Attacker input reaches the affected intrinsic via Jenkins's Base64 processing.
- This is a demonstration where the same flaw is chained into RCE using Jenkins-specific steps, separate from Path A.
Path C: Keycloak
- Attacker input reaches the affected intrinsic via Keycloak's Base64 processing.
- This is a demonstration where the same flaw is chained into RCE using Keycloak-specific steps, separate from Paths A and B.
Memory corruption can lead to crashes or information disclosure, but it does not lead to RCE via the crash itself.
4. Attacker Position and Execution Location
A remote attacker capable of sending Base64 data to the target application. Code executes with the privileges of the affected JVM process.
5. Perspective of Victims and Administrators
Inference: The following outlines expected observations during an investigation and does not represent published observation results from victim environments. This depends on logging and monitoring configurations.
Users may experience service disruption during authentication or request processing. Administrators may observe JVM crashes, SIGSEGV signals, hs_err logs, crafted requests, and Java child processes.
6. Success and Failure Conditions
The x86-64 CPU must feature avx512f, avx512bw, avx512vbmi, and bmi2; the affected JVM intrinsic must be enabled; and attacker input must reach the target Base64 decoding path. RCE requires additional conditions that combine product-specific heap layout control with information disclosure, and Base64 input alone is not guaranteed to succeed. ARM64 systems are not known to be affected. As a countermeasure, update to the patched version and restart all JVMs. According to CERT-EU, the fixed OpenJDK releases are 8u513 for the Java SE 8 performance build, 17.0.21.1 for 17, 21.0.13.1 for 21, 25.0.5.1 for 25, and 27.0.1.1 for 27. The affected range for 17 starts at 17.0.6. A temporary workaround, verified by VulnCheck on OpenJDK 17, 21, and 25, is using the startup options -XX:+UnlockDiagnosticVMOptions -XX:-UseBASE64Intrinsics or -XX:UseAVX=2. Verify on the target build and restart the JVM. The latter option is for x86-64 and must not be applied to ARM.
7. What Happens Upon Success
A JVM crash resulting in a DoS, or arbitrary code execution with the privileges of the application process, may occur. The research demonstrations establish feasibility; they do not constitute evidence of actual exploitation by third parties.
8. Observable Logs
Inference: The following lists potential log sources and artifacts for investigation. Because they depend on logging configurations, permissions, and retention periods, the absence of records does not prove an absence of compromise.
- Proxy/SWG/Web Access Logs: HTTP requests containing anomalous Base64 input (confirming the request content requires decryption or logging of the relevant fields; DNS logs cannot confirm HTTP request content)
- Endpoint/EDR: JVM crashes, core dumps, hs_err logs, Java child processes
- Identity/IdP: Anomalies immediately before or after Basic, JWT, or SAML processing
- Network: Repeated requests from the same source immediately preceding a crash
9. Attack Success Determination
Confirmed in Public Information
Calif reported demonstrating pre-authentication RCE in a controlled environment for Path A (Jira), Path B (Jenkins), and Path C (Keycloak). The publicly available standalone PoC only confirms memory corruption and does not indicate the release of product-specific RCE exploits or actual exploitation by third parties.
Criteria for Internal Assessment
For all three paths, crafted requests alone should be treated as attempts. Confirm causality between requests and crashes to determine a DoS, and substantiate execution of attacker-specified code to determine RCE. Do not conclude compromise based solely on Java child process generation or file writes; verify differences from normal processing and relations to exploitation. Furthermore, do not directly map public demonstrations to a compromise within your own organization.
10. Investigation Playbook
- Inventory JVM versions, CPU features, JVM flags, and Base64 reachability paths.
- Update to the vendor-patched version and restart the JVMs.
- Correlate crash logs with request logs to investigate child processes and file modifications.
- If signs of exploitation are found, isolate the host, verify credentials, and rebuild.
11. Defense and Detection Ideas
Add CPU features and JVM flags to asset data alongside software versions. Rather than relying solely on WAF patterns, correlate JVM crashes, child processes, and request sources for detection.
12. Facts / Inference / Hypothesis
- Fact: A memory corruption issue in the Base64 intrinsic on x86-64 CPUs with specific features.
- Fact: Multiple JDK families are affected, and patched versions and temporary flags have been released.
- Fact: Calif reported demonstrating pre-authentication RCE on Jira, Jenkins, and Keycloak in a controlled environment. The publicly available standalone PoC is intended for confirming memory corruption.
- Inference: Internet-facing Java services handling authentication or tokens are high-priority targets for review.
- Hypothesis: None.
13. MITRE ATT&CK Mapping
| ID | Technique | Confidence | Basis |
|---|---|---|---|
| T1190 | Exploit Public-Facing Application | High | Can lead to RCE from Base64 input paths in public Java applications. |
14. Unknowns and Additional Investigation
Actual exploitation has not been confirmed, and application-specific reachability and heap layout require individual verification. For vendor build patch statuses, the fixed OpenJDK release numbers (see Section 6) have been confirmed, but other vendor builds still require individual verification.
15. Impact on SOCs and Organizations
JREs bundled with products may not be updated solely through OS package updates. Servers meeting CPU conditions and internet-exposed authentication paths should be prioritized, and tracking should continue until restarts are completed, including for vendor appliances.
16. Summary by Role
- SOC: Correlate JVM crashes, crafted Base64 requests, and Java child processes.
- Administrators: Update to vendor-patched versions, restart all JVMs, and disable intrinsics if unpatched.
- Users: Report unexpected stoppages of Java services to administrators.
Top comments (0)