
Java4Shell is a critical flaw in Java that could let attackers run their own code on servers without logging in, and anyone running Java on an affected server should patch now. Oracle released a fix on 17 November 2026, and the US cyber agency CISA has urged organisations to apply it immediately.
The bug, tracked as CVE-2026-114938, was found by the security firm Calif with the help of artificial intelligence. It sits in the part of Java that decodes Base64, a common way of turning data into text, according to Calif’s own write-up. The BBC reported that it had been hiding in Java’s memory functions for more than a decade.
What is Java4Shell?
Java4Shell is a memory-corruption vulnerability in OpenJDK, the open-source Java runtime that many vendors build on. Calif says malformed Base64 input can overwrite neighbouring memory, which an attacker can use to read sensitive data or to take control of a system. The trigger can be as short as 64 letter As followed by an equals sign.
Java is the world’s third most popular programming language, according to Calif, which cites the TIOBE index. It quietly runs banks, hospitals, airlines and government systems. Most people never install it deliberately, which is partly why this one matters: it can sit inside other products.
Calif rated the flaw 9.2 out of 10 (critical) on the newer CVSS 4.0 scale. Oracle scored it lower, 8.1 (high), on CVSS 3.1, as BleepingComputer reported. Both scores put it firmly in the “fix this week” category.
Who is affected by the Java flaw?
Not every Java installation is exposed. Calif says the flaw affects OpenJDK 17.0.6 and later, and that it only bites on x86-64 processors that support a feature called AVX-512 VBMI. That covers recent Intel Xeon server chips and AMD’s Zen 4 and newer. Chips based on ARM are not affected.
Because the code comes from OpenJDK, builds from other vendors are caught too, among them Amazon Corretto, Eclipse Temurin, Microsoft’s build of OpenJDK, Azul Zulu and Red Hat. Calif demonstrated remote code execution without a login against Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate in their default set-ups. It also showed attacks that need a login against products including Elasticsearch and Apache Kafka.
There is an important caveat. Calif says there is no universal exploit: each target needs its own tailored attack, and the researchers have published a proof of concept for the bug itself but are holding back the weaponised chains until people have had time to update.
How to patch Java4Shell
Oracle’s fix arrived in its November 2026 Critical Security Patch Update, which covers 655 vulnerabilities across its products. CISA’s alert tells organisations to find every Java installation, including runtimes bundled inside other software, container images and appliances, and to patch internet-facing systems first.
Where patching has to wait, Calif suggests two temporary workarounds. Starting Java with the option -XX:-UseBASE64Intrinsics switches off the vulnerable accelerated decoder. On x86-64 machines, -XX:UseAVX=2 turns off AVX-512 altogether. The researchers stress these are stop-gaps, not a substitute for updating.
What it means for you
If you are an ordinary home user, the risk is low. The bug matters mainly for servers, and the vulnerable code only runs on certain processors. You do not need to panic, but it is sensible to keep your computer’s software updates switched on and to update Java if you installed it yourself.
If you work in IT, or your employer runs tools such as Jenkins, Jira or Confluence, ask your IT team whether the Java behind them has been updated. Small firms using managed services should ask their supplier directly and get an answer in writing.
It also lands in a week when Britain has been asking how much it depends on one supplier after the Microsoft outage. Java is a different kind of dependency, buried in countless products, but the lesson is similar: shared foundations carry shared risk.
Why this one was found by AI
Calif says its researcher Quang Luong found the flaw in about an hour using OpenAI’s Codex coding assistant, and built a first working exploit in roughly two more hours, at a total cost of under $200. The firm earlier disclosed a similar flaw in PHP. Computer scientist Professor Toby Walsh told the BBC that AI is now exposing flaws “much more quickly”, including ones humans might never have found.
The same speed cuts both ways. Defenders can use AI to audit code, but attackers can too, which shrinks the gap between a bug being disclosed and being used. That is why agencies are pushing for rapid patching rather than the usual monthly rhythm.
For more on how artificial intelligence is reshaping the industry, see our guide to the Anthropic IPO and the rest of our technology coverage.
What happens next
Other OpenJDK vendors are still rolling out their own fixes, so check your supplier’s advisory for the exact build number. Calif has promised to publish its product-specific exploits later, once users have had time to upgrade. Expect more vendors to issue guidance over the coming days, and treat any unpatched, internet-facing Java server as the first job on Thursday morning.
































