A serious security flaw in Java, nicknamed “Java4Shell”, can let attackers take over some widely used business software without logging in. Oracle and the OpenJDK project released fixes at 1pm Pacific time on Tuesday 17 November, which was 8am Wednesday in Sydney, according to the researchers who found it, security start-up Calif. The US cyber agency CISA says it “strongly urges organizations to immediately patch the recently disclosed Oracle Java vulnerability CVE-2026-114938.”
Most Australians will never touch this directly. But Java runs a huge share of the systems behind banks, insurers, government agencies, universities and online services, including popular tools such as Atlassian’s Jira and Confluence. Here is what the flaw is, who is affected, what the fixes are, and what Australian organisations should do now.
What Java4Shell is
The bug, tracked as CVE-2026-114938, is in the part of Java that decodes Base64, a common way of turning data into text for logins, file uploads and network messages. Calif says that on certain processors OpenJDK uses specially optimised machine code to decode Base64 faster, and a mistake in that code lets malformed input overwrite neighbouring memory. Calif says the corruption can be triggered with just 64 “A” characters followed by an equals sign.
In the worst case, that can lead to remote code execution, meaning an attacker runs their own code on the server. Red Hat’s security bulletin says an attacker who can supply malformed Base64 input to an affected application “can take full control of the application or crash the service.” Calif rates the flaw 9.2 out of 10 (critical) under the CVSS 4.0 scoring system. Oracle’s November advisory scores it 8.1 (high) under CVSS 3.1, with high attack complexity, and says it may be exploited over a network without a username or password.
Who is affected
Three conditions have to line up, according to Calif, Red Hat and Datadog Security Labs:
- A recent Java version. OpenJDK 17.0.6 and later, including Java 17, 21, 25 and 27 builds before the fixes. Red Hat says: “Customers running OpenJDK 8 or OpenJDK 11 are also not affected, as these versions do not contain the vulnerable code path.” Oracle also lists one specialised Java 8 build, 8u511b33-perf.
- A particular kind of processor. x86-64 chips with AVX-512 VBMI instructions, which Calif says includes Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids and Granite Rapids, and AMD EPYC and Ryzen chips from Zen 4 onwards. Calif says ARM64 processors are not affected.
- An application that decodes Base64 from outside. That is common. Red Hat notes that login credentials, single sign-on data and security tokens are often Base64-encoded and decoded before a user has finished logging in.
The list of vulnerable Java builds is long because almost every Java distribution is built from OpenJDK. Calif names Amazon Corretto, Eclipse Temurin, the Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine, Red Hat OpenJDK, Oracle OpenJDK, Oracle Java SE and Oracle GraalVM for JDK.
Which apps were shown to be exploitable
Calif says its testing achieved remote code execution in default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate without logging in, and in Apache Kafka, Apache Solr, Elasticsearch, Gerrit and Spinnaker after logging in. It says the lists are not exhaustive. These are Calif’s own results; Tenable and other analysts repeat them, but we have not seen the product-specific exploits independently reproduced, because Calif is holding them back “until everyone has had a chance to upgrade.”
Calif also stresses that “there is no universal exploit”: an attacker has to build a separate attack for each application. Datadog says crashing a vulnerable service is the more readily achievable impact, while reliable code execution “requires additional exploitation work tailored to the application”. Calif has published a proof-of-concept that triggers the memory corruption but does not take over a system.
Is anyone being attacked yet?
Not that anyone has reported. Tenable said that as of 17 November it was not aware of any exploitation, and that the flaw was not on CISA’s list of known exploited vulnerabilities. Canada’s Cyber Centre said: “The Cyber Centre is not aware of active exploitation at this time, but that can change quickly.” Cisco Talos, quoted by The Register, said it “has not observed in-the-wild exploitation of this vulnerability yet. However, we anticipate probing attempts may pick up soon after public disclosure.”
What to update to
According to Oracle’s release information as summarised by Datadog, the first fixed releases are Java 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. The OpenJDK advisory lists the flaw as affecting Java 17, 21, 25 and 27. Other vendors are shipping their own builds; Red Hat says it is expediting fixes for Red Hat Enterprise Linux 8, 9 and 10 and its OpenJDK 17, 21 and 25 builds.
Where patching has to wait, Calif, Red Hat and Datadog describe two temporary workarounds when starting Java: -XX:-UseBASE64Intrinsics, which turns off only the faster Base64 code and is Red Hat’s preferred option, or -XX:UseAVX=2, which turns off AVX-512 entirely. Both warn that the second flag must only be used on x86-64 machines, because Java on ARM does not recognise it and will fail to start.
What it means for Australia
As of Wednesday afternoon we had not found a Java4Shell-specific alert on cyber.gov.au. On 17 November the Australian Signals Directorate’s Australian Cyber Security Centre (ASD’s ACSC) did publish a general critical alert telling organisations to speed up patching, warning that “Some vulnerabilities were also observed being successfully exploited within 48 hours of patches or mitigation advice becoming available.” The page does not name Java4Shell. Australia’s AUSCERT has published a members-only bulletin titled “Java4Shell and OpenJDK”, and New Zealand’s NCSC has issued a public alert.
The Australian exposure is likely to sit with the IT teams that run self-hosted Java software: Jira and Confluence servers, Jenkins build systems, Keycloak single sign-on, search and data platforms, and the many in-house business systems written in Java. Java runtimes are also often bundled inside vendor products, appliances and container images, where they may not show up in a normal software inventory, as CISA points out. For cloud-hosted versions of these tools, the provider runs the Java runtime, so check its status page or advisories rather than assuming.
Practical steps for Australian organisations, drawn from the CISA, Cyber Centre, NCSC and vendor guidance:
- Find every Java runtime, including those inside third-party products and containers, and note its version.
- Update to a fixed build from your Java vendor, starting with internet-facing systems, then restart every Java process so the new version is actually running. Rebuild container images that bundle Java.
- Check whether servers use affected processors. Calif publishes a Linux command for this; CPUs older than Intel Ice Lake or AMD Zen 4 are not known to be affected, according to Red Hat.
- If you cannot patch straight away, apply one of the JVM workarounds above or restrict access to exposed systems, and watch for unexpected Java crashes.
- Ask your IT provider or software vendors whether their products bundle an affected Java build and when they will ship updates.
For home users, there is no sign that ordinary PCs and phones are the main target. If a desktop program asks to update its bundled Java, accept it.
The AI angle
Calif says it used AI to find and weaponise the bug. Its researcher Quang Luong writes that he directed OpenAI’s Codex to audit OpenJDK’s internals, and Calif says it took him an hour to find the flaw and two more hours to build a first working exploit. “The total token cost was less than $200,” the company says. Gizmodo, citing the New York Times, reports the work used OpenAI’s GPT-6 Cyber model. Oracle’s advisory credits Luong of Calif.io and a second researcher, Albert Yang, for reporting the flaw.
Not everyone is impressed with how it was announced. Akamai’s Jan Schaumann, quoted by The Register, wrote on the Openwall mailing list: “Calif.io set their priorities on publicity-maxxing, it seems.” Calif is the same firm behind October’s PHP2Shell disclosure, which we covered in our guide to checking whether a WordPress site was safe from PHP2Shell.
What we don’t know yet
- Whether attackers will build working exploits for specific products, and how quickly. Calif says it will publish its own product-specific chains later.
- Which commercial products bundle an affected Java build. Vendors are still publishing advisories, and we found no Atlassian advisory for Java4Shell on its security advisories page at the time of writing.
- Whether ASD’s ACSC will issue Java4Shell-specific advice.
Our read
This is analysis, not reporting. Java4Shell is not as simple to exploit as Log4Shell, because it needs specific chips and a bespoke attack for each app. But the set of affected systems is huge, the trigger is trivial, and the fact that a small team with AI tools found and weaponised it in hours is exactly the speed-up ASD’s ACSC warned about the same day. Combined with the FBI’s recent warning about AI hacking frameworks like ARTEX, the safe assumption for Australian IT teams is that the time between disclosure and real attacks is shrinking, so this is one to patch this week, not next quarter.
Featured image: Oracle’s Redwood Shores campus in California, photographed in November 2009 (file photo). Photo: “Oracle Headquarters Redwood Shores” by Håkan Dahlström via Wikimedia Commons, licensed under CC BY 2.0. Resized to 1200×800 and converted to PNG. Oracle and Java are trademarks of Oracle; no endorsement implied.




