The US Cybersecurity and Infrastructure Security Agency (CISA) has urged organizations to immediately patch a critical Java vulnerability that could let attackers take over servers without a password.
The flaw, tracked as CVE-2026-114938 and nicknamed “Java4Shell,” was disclosed on Tuesday by security firm Calif.
Oracle and OpenJDK released fixes the same day.
Calif rated the flaw 9.2 out of 10 on the CVSS scale. No exploitation in the wild had been reported as of Tuesday.
The vulnerability was found in code Java uses to decode Base64, a common way of sending data as text.
A short, malformed Base64 string causes the decoder to write past the end of the memory set aside for it, Calif said. Calif said an attacker could use that memory corruption to read sensitive data from the server, such as keys and tokens, and then take full control of it.
The flaw affects OpenJDK 17.0.6 and later, and the Java distributions built on it, including Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Red Hat OpenJDK and Oracle Java SE.
It only affects x86-64 processors that support AVX-512 instructions, which include recent Intel Xeon and AMD EPYC server chips. ARM-based systems are not affected.
What to do
CISA urged organizations to identify every Java installation, including runtimes bundled inside third-party applications, container images and appliances, which may not appear as separate entries in software inventories.
Organizations should patch as soon as fixes are available, starting with internet-facing systems. Where that is not possible, CISA recommends restricting and validating Base64 input, enforcing input-size limits and limiting unauthenticated access, while watching for unexpected JVM crashes and suspicious Java processes.
Calif said the flaw can also be mitigated by starting Java with the -XX:-UseBASE64Intrinsics flag, which turns off the vulnerable hardware-accelerated decoder.
Calif researcher Quang Luong found this vulnerability in about an hour by using OpenAI’s Codex to search for vulnerabilities in OpenJDK’s code.
The find comes a day after Google Threat Intelligence Group reported attackers using AI agents to turn security patches into exploits without human input, shrinking the time organizations have to patch.