Oracle and OpenJDK patch high-severity Java flaw
Oracle and OpenJDK patched CVE-2026-114938, an x86-64 Java memory-corruption flaw that can enable unauthenticated remote code execution under specific conditions. Calif says AI assisted its discovery, but its public disclosure names Codex rather than GPT-6 Cyber.
Oracle and the OpenJDK project released fixes on November 17 for CVE-2026-114938, a high-severity Java memory-corruption flaw that can allow remote code execution without authentication. Operators need to replace affected runtimes and restart their JVMs before running processes use the fix.
Specific Oracle Java SE, OpenJDK and GraalVM releases are affected, and the vulnerable path runs on x86-64 processors with AVX-512F, AVX-512BW, AVX-512VBMI and BMI2 support. Exploitation also requires the optimized Base64 decoder to be active and attacker-controlled Base64 data to reach it. ARM64 systems are not affected by this hardware-specific path.
Oracle assigned a CVSS 3.1 score of 8.1 to the Java Compiler issue. The vulnerable HotSpot routine can miscalculate the output length when processing malformed Base64 padding, then write beyond the decoded output buffer. Depending on the application and memory layout, successful exploitation can crash a service, expose heap contents or run arbitrary code in the affected process.
CERT-EU says potentially exposed data paths include HTTP Basic authentication, JSON Web Tokens, SAML messages, cookies and TLS data. An attacker still needs a reachable decoder route and application-specific conditions that turn the out-of-bounds write into a working exploit.
Affected versions include Oracle Java SE 8 performance build 8u511-perf; Java SE or OpenJDK 17 from 17.0.6 through 17.0.21; Java 21 through 21.0.13; Java 25 through 25.0.5; and Java 27 through 27.0.1. Some Oracle GraalVM JDK 17 and 21 releases are also affected.
For Oracle users, the listed fixes are 8u513-perf, 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. Oracle’s JDK 17.0.21.1 release notes say that build includes security fixes from the November update. Other OpenJDK distributors and products that bundle Java may use different fixed build identifiers. CERT-EU advised installing the applicable vendor update, rebuilding containers or bundled runtimes where needed, and restarting every affected JVM.
If immediate patching is not possible, CERT-EU and Calif’s technical disclosure point to two temporary mitigations: disabling the Base64 intrinsic or limiting AVX to level 2. Both require a JVM restart, and limiting AVX can reduce performance.
Calif credits researcher Quang Luong, working with AI, with the discovery and exploit development. Its disclosure says Codex surfaced the Base64 memory-corruption finding and that the team found the bug in about one hour, then produced a first working exploit in about two more hours at a token cost below $200.
Calif’s public write-up does not identify OpenAI’s GPT-6 Cyber as the model used for this finding. OpenAI’s model documentation describes the restricted model as intended for vulnerability investigation, reproduction in authorized test environments and fix verification, but does not tie it to Java4Shell.
Canada’s Cyber Centre said on November 17 that it was not aware of active exploitation. The reviewed evidence does not establish whether exploitation began after that cutoff.
More news

NewCore says Grok Bot kept attempting data reads after check-ins were declined

OpenAI releases CSAM image classifier, adds child-safety interventions

AMD introduces Versal devices with silicon-backside attack protection
