Oracle and the OpenJDK project released fixes on November 17 for CVE-2026-114938, a memory-corruption flaw in Java’s Base64 decoder that researchers at security firm Calif say they turned into unauthenticated remote code execution in widely used Java applications. The Cybersecurity and Infrastructure Security Agency issued an alert the same day urging organizations to patch immediately.
Calif, which calls the bug Java4Shell, said its researcher Quang Luong found it in about an hour after directing OpenAI’s Codex to audit OpenJDK’s HotSpot virtual machine, and built the first working exploit in two more hours. The total token cost was less than $200, the firm said. Calif said it first built pipelines and strategies for AI agents to use, and that afterward “most of the later chains are built almost autonomously with AI.”
The flaw sits in hand-written AVX-512 code that HotSpot uses on x86-64 processors to decode Base64 faster. According to Calif’s write-up, an input of 64 “A” characters followed by a single equals sign makes the routine compute a negative output length and write 62 bytes past the end of a 50-byte array. Java then rejects the input, but neighboring objects may already be corrupted.
The bug requires x86-64 chips with AVX-512 VBMI support, which Calif says include Intel’s Ice Lake, Sapphire Rapids, Emerald Rapids and Granite Rapids Xeons and AMD’s Zen 4 and newer processors. ARM64 processors are not affected.
Calif said its testing showed pre-authentication code execution in default configurations of Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate, and post-authentication execution in Apache Kafka, Apache Solr, Elasticsearch, Gerrit and Spinnaker. Ordinary features such as authentication and file uploads can expose the vulnerable decoder, the firm said. It also said there is no universal exploit, since each application needs its own approach.
Calif rates the bug 9.2 out of 10 under CVSS 4.0, while Oracle and OpenJDK score it 8.1 under CVSS 3.1, with high attack complexity. Oracle lists it as remotely exploitable without authentication, and OpenJDK’s advisory marks the 17, 21, 25 and 27 release lines as affected. Oracle’s list of affected versions includes Java SE 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3 and 27.0.1.0.3, along with GraalVM releases. Calif says distributions built from OpenJDK, including Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu and Red Hat OpenJDK, are also vulnerable.
Oracle’s November 2026 Critical Security Patch Update, which carries the fix among 655 new security patches, credits two reporters for the flaw: Luong and a researcher named Albert Yang. Calif says starting Java with the -XX:-UseBASE64Intrinsics flag or, on x86-64 hosts only, -XX:UseAVX=2 also mitigates the bug. Where immediate patching is not feasible, CISA recommends temporary controls such as restricting and validating Base64 input at the application boundary, enforcing input-size limits and limiting unauthenticated access.
Calif said it used AI to find and exploit Java4Shell as it did with PHP2Shell, the two PHP remote code execution flaws that the PHP project fixed in October. On November 16, Google Threat Intelligence Group said it had found an attacker framework that compares code before and after patches, then drafts and runs exploits without human involvement. Calif is withholding its product-specific exploit chains until users have had a chance to upgrade and says it will publish them later.
–
By the Control Plane Editorial Team