Oracle and the U.S. Cybersecurity and Infrastructure Security Agency (CISA) are urging organizations to patch Java runtimes immediately after researchers disclosed a critical flaw in OpenJDK’s HotSpot virtual machine. The bug, tracked as CVE-2026-114938 and branded “Java4Shell,” can let an attacker execute code on a server without ever logging in — and in some configurations, can also crash the Java Virtual Machine or leak sensitive data straight out of its memory.
Not Another Log4Shell
The name invites comparison to 2021’s infamous Log4Shell bug, but the two are unrelated. Log4Shell lived inside the popular Log4j logging library — something teams could often patch by updating a dependency. Java4Shell instead sits inside HotSpot itself, the JVM that actually executes Java bytecode across most OpenJDK-based distributions. That distinction matters operationally: simply bumping a library version in an application won’t fix anything here. Teams need to identify the actual Java runtime build each service uses, including any copy bundled by a third-party vendor inside an appliance or container image.
Security firm Calif, whose researcher Quang Luong found the bug while examining HotSpot with the help of an AI coding assistant, rated the issue 9.2 under the newer CVSS 4.0 standard — squarely critical. OpenJDK’s own advisory lists it at 8.1 under the older CVSS 3.1 scale. The two scores use different methodologies, but both agree this is a serious flaw, even though neither implies that every Java server everywhere is automatically exploitable.
Inside the Bug: A Base64 Decoder Gone Wrong
At the technical core is Base64 encoding — the common scheme used to represent binary data as plain text, heavily relied on in login flows, file uploads, and network protocols. On processors that support a CPU feature called AVX-512 VBMI (found in some newer Intel Xeon and AMD chips), HotSpot uses an optimized decoder that processes several bytes simultaneously for speed. A flaw in that fast path allows the decoder to write past the end of its allocated memory buffer before Java recognizes the input is malformed. By the time the bad input is rejected, the memory damage has already happened.
Researchers compared the resulting memory exposure risk to Heartbleed, since a successful attack could leak data already sitting in the process’s memory — session tokens, private keys, and other secrets — even though the underlying causes of the two bugs are completely different. Notably, ARM64 systems don’t use this particular code path and aren’t affected.
Who’s Actually at Risk
Calif’s testing found unauthenticated remote code execution was possible in default test configurations of several major products, including:
- Keycloak
- Atlassian Jira and Confluence
- JFrog Artifactory
- Jenkins
- PingFederate
Other tested products — including Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Spinnaker — required authentication first, somewhat lowering their real-world risk profile. Researchers also demonstrated exploitation in sample applications built with the Apache Struts and Spring Boot frameworks, and even showed a client-side attack path, where a vulnerable Java client could be compromised simply by opening a TLS connection to a malicious server. Security firm watchTowr noted there’s no single universal exploit; successful attacks depend on the specific combination of CPU, JDK build, and how an individual application happens to use Base64 decoding.
Fixes and Interim Mitigations
Oracle’s patched releases include Java SE 17.0.21.1, 21.0.13.1, 25.0.5.1, and 27.0.1.1, along with an updated Java 8 Enterprise Performance Pack build and fixes for several GraalVM editions. Organizations running other distributions — Amazon Corretto, Eclipse Temurin, Azul Zulu, and similar — should check their vendor’s own advisory for the equivalent fixed build.
CISA recommends patching internet-facing systems first, while specifically checking for Java runtimes bundled inside third-party software, container images, and network appliances, since these often don’t show up in a standard software inventory. Where immediate patching isn’t possible, Calif suggests two temporary JVM startup flags: -XX:-UseBASE64Intrinsics disables the vulnerable fast-path decoder outright, while -XX:UseAVX=2 turns off the AVX-512 instructions the bug depends on (this second flag should never be applied on ARM systems, where it can prevent Java from starting at all).
CISA also advises restricting unauthenticated access where feasible, validating Base64 input before it reaches the decoder, enforcing input-size limits, and watching closely for unexpected JVM crashes or unapproved Java runtimes — treating any unexplained crash on an exposed, unpatched system as worth investigating. After updating, administrators should restart or redeploy every affected service and verify the actual runtime version in use, rather than trusting the host’s default Java installation.
Leave a Reply
You must be logged in to post a comment.