It’s Tuesday in the security industry, which means thousands of organizations are once again conducting a large-scale, effort-intensive response to an embargoed vulnerability with high prevalence, high drama, and comparatively low exploitability. No, it’s not PHP2Shell: This time it’s “Java4Shell,” a high-severity (or critical if you’re using CVSS v4) memory corruption vulnerability in OpenJDK whose existence “leaked” predominantly because these high-hype, low-exploitability vuln drops teased in advance tend to look a lot more like marketing exercises than sincere attempts at risk-based analysis.
But much like with PHP2Shell, that doesn’t mean the vulnerability isn’t real, or that it isn’t exploitable for RCE…with a bunch of work.
Background
CVE-2026-114938 is an integer underflow vulnerability that leads to an out-of-bounds heap write in the HotSpot AVX-512 Base64 decode intrinsic in OpenJDK 17 (generate_base64_decodeBlock, reached via java.util.Base64.Decoder). The vulnerability arises from the intrinsic's padding handler decrementing a decoded-size counter with no lower bound: a crafted input larger than 128 bytes that ends in a short "=" padded tail drives the counter below zero and corrupts the destination pointer and store mask, thus producing a ~62-byte heap overflow past the decode buffer.
Per the official advisory, CVE-2026-114938 affects OpenJDK 17, 21, 25, and 27. According to a write-up from Calif, the following distributions built on OpenJDK are also vulnerable, in addition to a wide range of Java applications like Atlassian Jira and Confluence, Jenkins, Keycloak, and more:
- Amazon Corretto
- Eclipse Temurin
- Microsoft Build of OpenJDK
- Azul Zulu
- BellSoft Liberica
- SAP SapMachine
- Red Hat OpenJDK
- Oracle OpenJDK
- Oracle Java SE
- Oracle GraalVM for JDK
The vulnerability is present in servers and cloud systems running on current-generation Intel server processors (e.g., the Xeon chips in Intel's Ice Lake and Sapphire Rapids generations) or recent AMD processors (the Zen 4 / Zen 5 generation — EPYC Genoa/Turin servers, and also Ryzen 7000/9000 desktops). It would not be present on older Intel server chips, on most modern Intel desktops and laptops (Intel has fused AVX-512 off on all of its client processors since the 12th-generation "Alder Lake" parts, including the current Core Ultra / Arrow Lake), or on non-Intel/AMD processors such as ARM-based chips (e.g., Amazon's Graviton and Ampere, common in some cloud setups). One exception is Intel's 2019–2021 client chips (Ice Lake / Tiger Lake / Rocket Lake), which do expose the feature.
CVE-2026-114938 is not known to be exploited in the wild at time of writing.
VulnCheck research observations
VulnCheck’s research team reproduced the vulnerability on Ice Lake hardware and determined that exploitation is hardware-specific. The vulnerable code runs only on CPUs that provide the AVX-512 VBMI (and BW) instruction features; CPUs without them fall back to a different decode path that does not contain the defect, so the vulnerable instructions don’t execute there.
Based on our team's early reproduction, triggering the memory corruption to crash an affected JVM (a DoS) is trivial — if you can find and reach a sink that uses the vulnerable Base64 decode logic — and requires no special exploitation mechanisms, exploit development, or heap grooming. The attacker only needs to send the crafted input repeatedly until the overflow lands on a live object. (A single request usually will not crash a default-configured server, because the JVM's thread-local allocation buffers absorb most of these overflows, but a simple burst of requests reliably brought the service down in our testing.)
Remote code execution, by contrast, has a substantially higher bar: Successful exploitation for RCE requires non-trivial heap grooming combined with a target-specific memory disclosure path in order to leak addresses and bypass ASLR. In other words, we don’t anticipate that CVE-2026-114938 will be easily or widely exploitable, but as with PHP2Shell, adversaries are likely to find narrow but functional paths to RCE in specific applications. Inevitably, throngs of researchers will also point AI at OpenJDK and start pumping out public PoCs that trigger the vulnerability but fail to defeat ASLR. Calif’s own PoC merely triggers the vulnerability.
VulnCheck exploit availability
A full RCE exploit for CVE-2026-114938 is available to Initial Access Intelligence customers as of 6 PM EST on November 17. Our exploit ships with a version scanner, a PCAP, and Snort, Suricata, Sigma, and YARA rules.
Mitigations
Oracle’s advisory for CVE-2026-114938 has a full range of supported affected versions here, which at time of writing encompassed 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3, and 27.0.1.0.3.
In our team’s testing, disabling the intrinsic via either of the following options stops the overflow, as verified on OpenJDK 17, 21 and 25:
-XX:+UnlockDiagnosticVMOptions -XX:-UseBASE64Intrinsics-XX:UseAVX=2
We’ll update this blog with further information about remotely exploitable attack vectors, exploitation in the wild, and VulnCheck research observations as we verify additional details.
About VulnCheck
VulnCheck empowers organizations to transcend the challenges of vulnerability prioritization. Our suite of solutions provides product managers, PSIRT teams, and threat hunters with the tools required for accelerated, high-precision operations and infinite efficiency.
Recognizing the industry-wide necessity for superior data velocity and accuracy, we deliver high-fidelity insights to the market. We remain committed to surfacing critical intelligence on vulnerability exploitation and emerging trends, leveraging our unique dataset to support the practitioner community.
To deepen your understanding of these threats, VulnCheck Exploit & Vulnerability Intelligence provides comprehensive coverage of global threat actors. Register for a demo to explore our intelligence today.