Wednesday, November 18, 2026 · Beirut, Lebanon About · Contact
Cybersecurity

Java4Shell OpenJDK flaw allows unauthenticated remote code execution

Malformed Base64 strings can corrupt memory on x86-64 servers running Java 17.0.6 and later, exposing widely used business software to pre-authentication takeover.

I
November 18, 2026 · 3 min read
Java4Shell OpenJDK flaw allows unauthenticated remote code executionCybersecurity

Oracle addressed the Java4Shell OpenJDK flaw, tracked as CVE-2026-114938, on November 17 as part of a patch update covering 655 security vulnerabilities. Testing showed that attackers without credentials could achieve remote code execution (RCE) on default installations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins, and PingFederate.

How the Base64 memory corruption works

The bug resides in java.util.Base64.Decoder, which processes encoded data in HTTP Basic credentials, SAML artefacts, CSRF tokens, file uploads, and MIME encoded words. Because servers routinely decode these inputs before completing login checks, an attacker without an account can reach the vulnerable code path.

On x86-64 processors supporting AVX-512 VBMI instructions, the Java Virtual Machine (JVM) selects hand-rolled assembly code for frequently run Base64 decoding routines. A 65-character input of 64 A characters and an equals sign triggers a negative length calculation, while older OpenJDK 17 builds require 128 A characters. The runtime allocates a 50-byte buffer and throws an IllegalArgumentException, but only after writing 64 bytes at offset 48 and overflowing by 62 bytes.

Attackers only control where those fixed overflow bytes land on the heap, so Calif arranged server memory to overwrite a neighbouring array's length. That step lets an attacker write chosen bytes and read back heap secrets like private keys and tokens, a technique called JavaBleed, before reaching RCE.

Affected platforms and AI discovery

Calif researcher Quang Luong found the bug in about an hour using OpenAI's Codex artificial intelligence (AI) coding assistant. He developed a first working exploit in two more hours at a total AI token cost of less than $200. AI agents also helped build additional exploitation chains, and Calif achieved authenticated RCE on Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Spinnaker. Calif compromised sample Apache Struts and Spring Boot applications too, while researcher samczsun shared a client-side exploit triggered over outgoing TLS connections.

The vulnerable code affects OpenJDK 17.0.6 and later, including Java 21, 25, and 27, Java SE 8 Performance Pack build 8u511b33-perf, and several GraalVM builds. Standard OpenJDK 8 and 11 and ARM64 systems are unaffected. Exploitation requires Intel Ice Lake, Sapphire Rapids, Emerald Rapids, or Granite Rapids processors, or AMD Zen 4, Zen 5, and Zen 6 chips.

Calif withheld its weaponised application exploits, and Canada's Cyber Centre reported no known active attacks this week. Security firm Upwind reproduced the memory corruption and JVM crashes on an AWS server running an Intel Ice Lake processor. Blackpoint Cyber's Adversary Pursuit Group crashed a test Spring Boot application by sending about 28,300 malformed login requests over roughly 20 seconds. The group also used the JavaBleed step in local tests to read a secret stored next to the overflow across all 30 runs.

What happens next

Scoring the bug at 8.1 under CVSS 3.1, Oracle said, "Successful attacks of this vulnerability can result in takeover of Oracle Java SE." Oracle released fixed builds 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1, and 8u513, while Amazon Corretto, Azul Zulu, and GraalVM also issued updates. Red Hat assigned a preliminary 9.8 score and is expediting fixes for Red Hat Enterprise Linux (RHEL) 8, 9, and 10 and RHEL container images.

Until administrators install patched runtimes and restart Java services, they can disable the vulnerable routine by starting the JVM with the -XX:-UseBASE64Intrinsics flag. Administrators on x86-64 hosts can alternatively pass -XX:UseAVX=2 to turn off AVX-512 instructions, though that flag stops ARM64 JVMs from starting. Defenders can also monitor logs for IllegalArgumentException bursts, hs_err_pid crash files, and malformed Base64 tokens of 32 characters or more in login headers.

Sources: BleepingComputer, Calif, Red Hat, The Hacker News

Fact-checked against the linked sources before publication. Spotted an error? See our Corrections Policy.

Spotted an error? Email corrections@itmagazineme.com with the article link.

Related stories