Patch Java now: AI helped researchers find a critical flaw in widely used Java versions

A critical Java flaw found with help from an AI coding agent for under $200 is now fixed; update Java 17, 21, 25 and 27 on x86-64 servers.

Under $200: of computing to find a critical Java flaw with an AI coding agent

Oracle and the OpenJDK project released fixes on 17 November for Java4Shell (CVE-2026-114938), a critical memory-corruption flaw in Java’s built-in Base64 decoder that can let an attacker run code on a server. Security firm Calif, which disclosed it, says researcher Quang Luong found the bug in about an hour by pointing OpenAI’s Codex coding agent at parts of the Java runtime, spending under $200 on computing, and built a first working exploit in two more hours.

The flaw affects Java 17 (from 17.0.6), 21, 25 and 27, including builds from Oracle, Eclipse Temurin, Amazon Corretto, Azul Zulu, Red Hat and others, but only on x86-64 processors with the AVX-512 VBMI feature, found in many recent Intel Xeon and AMD server chips. ARM machines are not affected.

Why it matters

Java runs inside widely used tools such as Jira, Jenkins and Keycloak, and researchers report some can be attacked without logging in. Tenable says exploitation attempts have been seen, though none confirmed as successful, and the flaw is not yet on CISA’s exploited list. CISA has urged organisations to patch immediately.

What to do: update to 17.0.21.1, 21.0.13.1, 25.0.5.1 or 27.0.1.1, or your vendor’s fixed build. If you cannot patch yet, Calif suggests starting Java with -XX:-UseBASE64Intrinsics. Read Calif’s disclosure.


Related