A critical memory corruption vulnerability in OpenJDK's Base64 decoder enables unauthenticated attackers to execute arbitrary code on vulnerable systems. Dubbed "Java4Shell" and tracked as CVE-2026-114938, the flaw was identified by security research firm Calif and patched this week by Oracle and the upstream OpenJDK project.

According to reporting from The Hacker News, the vulnerability resides within an optimized decoding path that accelerates Base64 processing using specialized vector instructions. The flaw impacts Java 17.0.6 and later versions across Java 17, 21, 25, and 27 releases, but its execution requires modern x86-64 processors equipped with AVX-512 VBMI capabilities.

Hardware Prerequisites and Memory Corruption Mechanics

The issue manifests when the Java Virtual Machine (JVM) compiles frequently executed code paths using AVX-512 VBMI instruction sets. An attacker can supply malformed Base64 data ending with an unexpected equal sign, causing the decoder to write 62 or 63 bytes beyond the intended output buffer before rejecting the malformed string. As detailed by Help Net Security, a trigger can consist of as little as 64 "A" characters followed by an equal sign in modern versions, or 128 characters in older OpenJDK 17 builds.

Because the overflow writes fixed bytes, weaponization requires precise heap grooming to alter adjacent memory structures. Calif demonstrated that expanding neighboring arrays allows arbitrary memory writes alongside an information leak technique dubbed "JavaBleed." This memory leak allows attackers to bypass address space layout randomization (ASLR) and gain complete control over execution flow.

Exploitation is strictly constrained by CPU architecture. Vulnerable instructions run exclusively on systems supporting AVX-512 VBMI, including Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids, Granite Rapids, and AMD Zen 4 or newer processors. ARM64 chips and older x86-64 hardware fall back to safe decoding routines and are unaffected.

Affected Applications and Enterprise Exposure

Attackers can reach the vulnerable decoder via common pre-authentication inputs, including HTTP Basic credentials, SAML artifacts, and Cross-Site Request Forgery (CSRF) tokens. Calif successfully demonstrated unauthenticated remote code execution against default installations of Atlassian Jira, Atlassian Confluence, Keycloak, Jenkins, JFrog Artifactory, and PingFederate.

Authenticated code execution was achieved in Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Spinnaker. Furthermore, security researcher samczsun uncovered a client-side vector where outbound TLS connections to malicious endpoints could trigger the flaw. While third-party security firms have independently replicated denial-of-service crashes and memory leaks, they note that crafting reliable remote code execution chains requires significant per-application tailoring.

Vendor Response and Severity Ratings

CVSS assessments vary across industry bodies. Calif assigned the vulnerability a CVSS 4.0 score of 9.2, while Red Hat issued a preliminary score of 9.8. Oracle assigned a CVSS 3.1 base score of 8.1, citing high attack complexity due to the precise heap manipulation required for code execution. Amazon rated the flaw at 8.8.

Canada's Cyber Centre and the Cybersecurity and Infrastructure Security Agency (CISA) have warned organizations regarding the exposure. Although CISA has not recorded active exploitation in its Known Exploited Vulnerabilities catalog, the agency strongly urges defenders to deploy vendor updates immediately.

Patches are available for Oracle Java SE versions 17.0.21.1, 21.0.13.1, 25.0.5.1, and 27.0.1.1, alongside updated GraalVM builds and Java SE 8 Performance Pack build 8u513. Amazon Corretto and Azul Zulu have released patched packages, while SAP has published updated SapMachine builds. Red Hat, Eclipse Temurin, BellSoft Liberica, and Microsoft Build of OpenJDK are working to finalize patched distributions.

What You Should Do

Security and infrastructure teams should immediately audit environments for vulnerable runtimes and compatible x86-64 hardware. Linux administrators can verify CPU exposure by executing:

grep -m1 '^flags' /proc/cpuinfo | tr ' ' '\n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'

If all four CPU flags appear, the host is capable of running the vulnerable routine. Organizations should implement the following defensive actions:

  • Apply runtime patches: Update OpenJDK and vendor-specific Java installations to patched releases, followed by restarting Java services and redeploying affected container images.
  • Deploy JVM mitigation flags: If direct patching is delayed, administrators can disable vulnerable code paths by passing the -XX:-UseBASE64Intrinsics flag to the JVM, disabling the vulnerable routine without disabling other AVX-512 features. Alternatively, -XX:UseAVX=2 can be used on x86-64 systems to turn off AVX-512 entirely.
  • Enforce boundary filtering: Inspect reverse proxies and web application firewalls to block Base64 strings longer than 32 characters ending with padding characters that violate standard four-character length multiples.
  • Monitor system telemetry: Review application logs for unusual spikes in java.util.Base64$Decoder IllegalArgumentException errors and investigate sudden JVM crashes marked by unexpected hs_err_pid log files.