Java4Shell OpenJDK flaw: Exploitation is tricky and hinges on specific CPUs (CVE-2026-114938)
Our summary
Researchers at Calif disclosed CVE-2026-114938, dubbed "Java4Shell", a memory-corruption bug in OpenJDK's Base64 decoder that stems from the optimized x86-64 machine code used to speed up decoding on CPUs supporting AVX-512 VBMI/BW instructions. On vulnerable hardware, malformed Base64 input as short as 64 'A' characters plus an equals sign can corrupt adjacent memory, potentially leaking data or, through careful heap grooming and an ASLR bypass, leading to unauthenticated remote code execution. Calif demonstrated unauthenticated RCE in default setups of Keycloak, Atlassian Jira, Confluence, JFrog Artifactory, Jenkins, and PingFederate, plus post-auth RCE in Kafka Broker, Solr, Elasticsearch, Gerrit, and Spinnaker, while apps built on Struts and Spring Boot are also at risk since any feature passing data through java.util.Base64.Decoder (auth headers, SAML, CSRF tokens) could be abused. Oracle fixed the flaw (CVSS 3.1: 8.1; Calif's CVSS 4.0 score: 9.2) across Java SE 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3, 27.0.1.0.3, GraalVM for JDK, and GraalVM Enterprise Edition, and downstream distributions like Corretto, Temurin, Zulu, SapMachine, Liberica, and Red Hat OpenJDK are rolling out fixes too. ARM64 systems are not impacted, no in-the-wild exploitation has been confirmed so far, and experts including VulnCheck and watchTowr expect exploitation to stay narrow and application-specific rather than widespread—but CISA still recommends inventorying Java deployments and patching without delay, or applying compensating controls like input-size limits and Base64 validation at the application boundary if patching must wait.
Help Net Security publishes this story on its own site; we link to it rather than reprint it.
Worried this affects your company?
Discuss a security assessment of your internet-facing systems. Scope agreed before testing.