Security firm Calif Research disclosed a critical vulnerability in the OpenJDK Java runtime on November 17, dubbed Java4Shell and tracked as CVE-2026-114938, scoring 9.2 out of 10 under CVSS 4.0. Oracle shipped patches the same afternoon at 1:00 p.m. Pacific as part of its November Critical Patch Update, and the OpenJDK project published its own advisory. The bug is a memory-corruption flaw in the hand-written AVX-512 assembly that OpenJDK uses to accelerate Base64 decoding — one of the most ubiquitous operations in server software, and one that very often runs on input straight from the network.
The mechanics are precise and uncomfortable. On x86-64 processors with AVX-512 VBMI support, the HotSpot JVM swaps in an optimized machine-code routine for Base64 decoding. A malformed 65-byte input — sixty-four "A" characters followed by a single "=" — makes the routine's length arithmetic underflow to negative, and a mask register then treats all 64 bytes as writable. A 50-byte output buffer receives a 64-byte write starting at offset 48, sending 62 bytes past its end into neighboring heap objects. On older Java 17 builds the trigger needs 128 "A" characters instead. Calif demonstrated that the corruption can leak heap memory Heartbleed-style — a primitive it named JavaBleed, potentially exposing keys and tokens — and escalate to full remote code execution.
The reach list is what turns this from an academic bug into a patch-today event. Calif showed pre-authentication RCE in the default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate, and post-auth RCE in Apache Kafka, Solr, Elasticsearch, Gerrit and Spinnaker. Sample applications built on Apache Struts and Spring Boot were also exploitable, and researcher samczsun contributed a client-side chain that compromises a Java program merely by having it open a TLS connection to a malicious server. There is no universal exploit — every target needs its own chain — and no exploitation in the wild has been reported.
The discovery process is the part the security industry will be quoting for months. Calif researcher Quang Luong directed OpenAI's Codex agent to audit the JIT compiler, its intrinsics and cryptography-adjacent code, found the vulnerability in about one hour, and had a first working exploit two hours later, for less than $200 in token costs. Calif framed the result as the arrival of the "centaur hacker" — part human, part AI — and noted that most early agent output was noise; the find came from a human steering the agent through it. The disclosure follows the firm's PHP2Shell finding earlier this month and lands in the same week Google Project Zero and Anthropic shortened their disclosure windows to 30 days on the argument that AI makes discovery cheap.
The affected scope is broad but bounded by hardware. Calif dates the vulnerable code to OpenJDK 17.0.6 and newer, and Oracle's advisory lists Java SE 8u511, 11.0.33, 17.0.21, 21.0.13, 25.0.5 and 27.0.1 as affected, with fixed builds 8u513, 11.0.33.1, 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. Distribution builds from Amazon Corretto, Eclipse Temurin, Microsoft, Azul Zulu, BellSoft Liberica, SAP SapMachine and Red Hat inherit the bug, and container base images and bundled JDKs are the copies operators tend to miss. Only x86-64 chips with AVX-512 VBMI are vulnerable — Intel Xeon from Ice Lake onward and AMD EPYC and Ryzen from Zen 4 onward — while ARM64 is unaffected. One scoring footnote: Oracle's own advisory rates its highest Java SE flaw this month at 8.1 under CVSS 3.1, versus Calif's 9.2 under CVSS 4.0.
Until fleets can patch, Calif published two workarounds: start the JVM with -XX:-UseBASE64Intrinsics to disable the accelerated decoder, or cap vector extensions with -XX:UseAVX=2, which only applies on x86-64 hosts. CISA issued a notice urging organizations to apply the patch, and South Korea's National Intelligence Service and Switzerland's National Cyber Security Centre both issued urgent inventory advisories. Oracle's November CPU covered 652 flaws across its product line in total, with 11 Java SE patches.
The structural lesson is the same one security teams have been absorbing all month, now with a sharper edge. A memory-safety bug in one of the most audited runtimes on earth, found in an hour for the price of a lunch, means the cost of finding is collapsing while the cost of shipping patches — and getting millions of deployments to install them — stays exactly where it was. That growing gap, not any single CVE, is the real problem Java4Shell leaves behind.
Comments (0)
Log in to join the discussion
Log InNo comments yet