Calif discloses AI-found Java flaw as security teams race to patch affected systems
Red Hat confirms a critical vulnerability limited to certain Java versions and server processors. Internet probing has been reported, but successful attacks remain unconfirmed.
Red Hat’s critical advisory identifies Java4Shell in OpenJDK’s Base64 decoder, with exposure limited to Java applications on x86-64 processors supporting AVX-512 VBMI when attacker-controlled input reaches the decoder. Calif says its AI-assisted search took about an hour and cost under $200; it demonstrated pre-login takeovers of default installations of several identity and business applications. GreyNoise saw probing but had not confirmed successful attacks, while Immersive says it built an independent exploit. Teams should inventory bundled runtimes and prioritize vendor-supported fixes or temporary mitigations.
01
Red Hat says OpenJDK 17.0.6 and later may contain the flaw; OpenJDK 8 and 11 lack the vulnerable code.
02
Red Hat’s preferred temporary workaround is `-XX:-UseBASE64Intrinsics`; `-XX:UseAVX=2` is an alternative only for x86-64 deployments.
03
Calif demonstrated password-free takeovers of default Keycloak, PingFederate, Jira and Confluence installations, but is withholding product-specific attack code.
An AI-assisted discovery has triggered an urgent security response for Java applications, including systems that handle workplace logins. Security startup Calif says it found the flaw in about an hour for under $200 in AI costs. Red Hat confirms that affected systems can let an attacker run code remotely without first logging in.
The vulnerability, CVE-2026-114938, is nicknamed Java4Shell. Internet monitoring company GreyNoise has detected requests probing for it, according to HRD’s November 18 coverage. GreyNoise had not confirmed successful attacks at the time of its report—a distinction between evidence that systems are being checked and evidence that they have been compromised.
A fast discovery, with working attacks already demonstrated
Calif says it used an AI coding tool to find the weakness. Its reported time and cost describe the discovery effort, not the time needed to repair every affected deployment. The company says it demonstrated password-free takeovers of default installations of Keycloak and PingFederate, used for single sign-on, along with Atlassian’s Jira and Confluence.
The potential damage extends beyond taking control. Calif says the flaw can expose information held in an application’s memory, potentially including personal information, passwords and security keys. It is withholding product-specific attack code for now. Security company Immersive, however, says it has independently built a working exploit.
The processor determines exposure
Red Hat’s November 17 advisory places the bug in OpenJDK’s Base64 decoder. Base64 is a way of representing data as text; applications use it for items such as login credentials and authentication tokens. Specially malformed input can corrupt memory, allowing an attacker to control the application or crash its service.
But this is not a warning that every Java installation is vulnerable. Red Hat identifies the affected code in OpenJDK 17.0.6 and later, including versions 21 and 25. Exploitation also requires an x86-64 processor supporting AVX-512 VBMI, a particular set of processor instructions, and an application that passes attacker-controlled data to the decoder.
Editorial illustration for Calif discloses AI-found Java flaw as security teams race to patch affected systems.
On those processors, Java automatically selects a hardware-accelerated decoding path unless it has been disabled. Red Hat says attacks before login have been demonstrated against widely deployed applications in their default configurations. Credentials and tokens are often decoded before authentication finishes, giving malicious input a route into the vulnerable code before a password is checked.
One flaw, multiple repair paths
Red Hat classifies the vulnerability as critical and says it is expediting fixes. Its affected-product list includes its OpenJDK 17, 21 and 25 builds and Red Hat Enterprise Linux 8, 9 and 10. Products and containers drawing Java packages from those systems may also need attention; updating one application does not inventory every bundled Java runtime.
While awaiting an update, Red Hat recommends disabling the accelerated Base64 implementation with this Java startup flag. It calls this the preferred temporary mitigation because other AVX-512 optimizations remain available.
-XX:-UseBASE64Intrinsics
An alternative flag, -XX:UseAVX=2, disables AVX-512 instructions entirely. Red Hat warns that this option belongs only on x86-64 deployments: ARM64 Java virtual machines do not recognize it and will fail to start. Neither workaround is required for systems outside the affected processor scope.
The vendor advises teams to identify each application’s Java supplier and full build version, including copies inside containers, appliances and product installations. Internet-facing applications and authentication services should take priority. For products with bundled Java, teams should use the product vendor’s supported update or mitigation path rather than replacing that runtime independently.
Container users have another step: rebuild their own containers after fixed base images become available. Red Hat says the startup flags are temporary measures, not substitutes for the permanent OpenJDK repair.
Sources
access.redhat.comRHSB-2026-018 Remote Code Execution via memory corruption on x86-64 systems - OpenJDK (CVE-2026-114938) - "java4shell" | Red Hat Customer Portal
hcamag.comEmployers told to patch critical Java flaw immediately as AI-found bug hits staff login systems
Reader comments
Newest comments first. Replies stay oldest first.