Java4Shell flaw in OpenJDK Base64 decoding exposes Jira, Jenkins and Keycloak
Researchers at Calif say a memory flaw in how OpenJDK decodes Base64 can let unauthenticated attackers run code on widely used enterprise software, though no exploitation has been reported in the wild.
By Mr Elite Published Updated 9 min read
- Severity
- High
- Status
- Reported by outlets
- Sources
- 5
- Last checked
Latest updates
- Help Net Security and Security Affairs report that Oracle and the OpenJDK project fixed the flaw on Tuesday (17 November), and give the exact trigger input and the cause: a padding calculation that underflows. Help Net Security adds Oracle's affected build numbers, SentinelOne's note on where payloads can arrive, a 'not world ending' remark from Kevin Beaumont, and CISA's interim safeguards.
The short version
- Calif published details of Java4Shell (CVE-2026-114938) on 17 November 2026 after some information leaked before the planned disclosure.
- Only x86-64 chips with AVX-512 VBMI are affected; ARM64 systems, including AWS Graviton, are not, according to Calif, VulnCheck and Endor Labs.
- Calif reported pre-authentication code execution on Keycloak, Jira, Confluence, Artifactory, Jenkins and PingFederate in default settings.
- Help Net Security reported that Oracle and the OpenJDK project fixed the bug on Tuesday 17 November; Endor Labs said other vendors' builds and public OpenJDK repositories had not caught up that day.
- Infosecurity Magazine reported that no attacks had been seen in the wild as of 18 November. VulnCheck and watchTowr do not expect indiscriminate exploitation.
What happened
Calif, a security research firm, published details on 17 November 2026 of Java4Shell, a flaw in OpenJDK that it says can be used to take over popular enterprise applications without logging in. Infosecurity Magazine reported that some information leaked before the coordinated disclosure date, which prompted the release. The Hacker News and SecurityWeek covered the disclosure, and Help Net Security, Security Affairs and Infosecurity Magazine have since added detail.
The flaw is tracked as CVE-2026-114938. Calif researcher Quang Luong found it in a routine that OpenJDK uses to decode Base64 (a way of writing binary data as plain text) on x86-64 processors with a feature called AVX-512 VBMI. Calif says malformed input can overwrite neighbouring memory. The trigger is 64 'A' characters followed by a single '=', or 128 'A' characters and an '=' on older OpenJDK 17.
Calif said the corruption can be used two ways. One is a memory leak it calls JavaBleed, which it compares to Heartbleed and which could expose sensitive memory such as keys and tokens. The other is remote code execution, meaning the attacker runs their own commands on the server.
In its tests, Calif reported pre-authentication code execution on Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate, all in default settings. It said code execution was also possible after authentication on Apache Kafka Broker, Solr, Elasticsearch, Gerrit and Spinnaker. Sample apps built on Apache Struts and Spring Boot were also compromised, and researcher samczsun shared a client-side exploit with Calif.
Calif's public proof-of-concept only checks whether a given JDK and CPU combination reproduces the memory corruption. The firm is holding back its product-specific exploit chains to give organisations time to upgrade. It said AI helped both to find and to exploit the bug: Security Affairs reported that Luong used OpenAI's Codex, took an hour to find the flaw and two more hours to build a first working exploit, at a token cost under $200.
Help Net Security reported that Oracle and the OpenJDK project fixed the flaw on Tuesday. Infosecurity Magazine reported that each Java vendor will ship its fixed build on its own schedule. CISA published an emergency advisory on 17 November urging organisations to patch, and Help Net Security reported there is no evidence yet of successful attacks.
Researchers have urged calm. Kevin Beaumont told Help Net Security the flaw is not world ending and advised patching as normal. VulnCheck and watchTowr do not expect indiscriminate exploitation.
Why it happened
According to Calif, the fault sits in an optimised Base64 decoding routine that OpenJDK uses on x86-64 CPUs. Security Affairs reported Calif's explanation: the decoder works out the output buffer size as the rounded-up input length divided by four, times three, minus the padding. When the input ends in padding characters such as '=', the code can compute a negative length, and that underflow turns into a full 64-byte write instead of the few bytes intended.
In Calif's worked example, a 65-character input makes OpenJDK allocate a 50-byte buffer, and the routine then writes 64 bytes starting at offset 48. Security Affairs described this as an out-of-bounds write past the end of the buffer, landing in neighbouring heap memory.
The attacker at first controls only where the data is written, not most of its content. Calif said it got round this by corrupting the length of a nearby array so that it could hold attacker-controlled data, then repeated the trick on a second array to read memory. Getting two objects to sit side by side in memory took careful timing of HTTP requests, and Calif rented a virtual machine in the same region as its target for a stable connection.
Calif says the decoder can be reached through everyday features such as authentication and file uploads. SentinelOne said any Base64 value an application passes to java.util.Base64.Decoder can carry the payload, including HTTP Basic credentials, SAML artifacts, CSRF tokens and MIME encoded words. Calif also reported a client-side route in which a Java application is compromised when it connects to an attacker-controlled server over TLS.
How this kind of failure works
Java is usually described as memory safe, because the runtime stops programmers from writing outside the space they were given. That promise holds for Java code. The runtime itself, however, contains native code and CPU-specific shortcuts written for speed, and a mistake in one of those can break the guarantee for every program running on top.
Picture a form with ten numbered boxes, where a clerk copies your answer into the boxes. If the clerk does not stop at box ten and keeps writing, the text spills into the next person's form. In a computer, the neighbouring data might be a password, a key, or a pointer that decides what code runs next. That spill is called an out-of-bounds write.
In this case the clerk first counts how many boxes are needed, and a subtraction goes wrong. Padding at the end of the input makes the count come out negative, and a negative count is read as a huge one. This kind of arithmetic slip, an integer underflow, is a common source of buffer overruns.
AVX-512 VBMI is an instruction set extension that lets a processor handle many bytes in one step, which makes decoding large Base64 strings faster. Runtimes often keep a fast path for chips that have such features and a slower path for the rest. That is why only some hardware is affected: the faulty code never runs on the slower path.
Turning a spill into code execution is harder than causing a crash. VulnCheck said an attacker must shape the heap layout and find a way to leak addresses and get past ASLR, which randomises where things sit in memory. That explains why it expects only narrow, application-specific paths rather than one exploit that works everywhere.
Who is affected
Calif says OpenJDK versions from 17.0.6 onward are vulnerable, along with distributions built on it: Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, SAP SapMachine, BellSoft Liberica, Red Hat OpenJDK and Oracle Java SE. Infosecurity Magazine reported that the flaw affects OpenJDK 17, 21, 25 and 27. Oracle's advisory, as relayed by Help Net Security, lists Java SE 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3 and 27.0.1.0.3 (and likely earlier), plus Oracle GraalVM for JDK and GraalVM Enterprise Edition.
Exploitation needs x86-64 hardware with AVX-512 VBMI, such as Intel Xeon chips from the Ice Lake generation onward and AMD chips based on Zen 4 or newer. Endor Labs noted these include recent server chips that account for much of current cloud capacity. ARM-based servers, including AWS Graviton, are not affected. VulnCheck said it reproduced the bug on Ice Lake hardware.
To check, find which Java version each server, container and appliance runs, and whether its CPU has the feature. On Linux, the CPU flags in /proc/cpuinfo list avx512_vbmi where it is supported. CISA warned that Java may be embedded in third-party software, containers and appliances and may not appear in asset inventories.
In practice, the worst case is a server that leaks secrets or runs an attacker's code. A more likely outcome is a crash that takes a service down. Oracle rates the flaw as high attack complexity, and Help Net Security reported Oracle's score as 8.1 on CVSS 3.1.
How to defend against it
Tick each step as you check it. Your progress is saved on this device.
What to learn from this
The concept: Shared runtime risk
Applications inherit the weaknesses of the platform under them. Jira, Jenkins and Keycloak did not each contain this bug. They all call the same Java decoder, so one flaw in it reaches all of them at once.
The same pattern appears with logging libraries, image parsers, compression code and TLS libraries. A single shared component becomes a single point of failure for thousands of products.
This is why a software inventory matters. If you cannot list where a runtime or library lives, including inside appliances and containers, you cannot tell how exposed you are when a flaw like this appears.
A flaw in a shared runtime is a flaw in every application that uses it, so know where your runtimes are.
Quick check
Sign in to earn 10 XP for a right answer
Why can one OpenJDK bug affect Jira, Jenkins and Keycloak at the same time?
The flaw lives in the Java runtime's decoder, not in each product. Any application that reaches that code on affected hardware inherits the problem.
Timeline
- Before 17 November 2026 Some information about the flaw leaks ahead of the planned coordinated disclosure, according to Infosecurity Magazine.
- 17 November 2026 Oracle and the OpenJDK project fix the flaw, according to Help Net Security.
- 17 November 2026 Calif publishes its findings and a proof-of-concept that shows the memory corruption. CISA publishes an emergency advisory.
- 17 November 2026 Endor Labs notes the fix is not yet in the public OpenJDK 17, 21 and 25 update repositories.
- 18 November 2026 Infosecurity Magazine and Help Net Security report no attacks in the wild; VulnCheck and Endor Labs publish analyses stressing the hardware dependence.
What is not known yet
- When Calif plans to publish the weaponised exploit chains for individual products.
- When all distribution vendors will ship patched builds.
- Whether attackers will find reliable code execution paths in specific applications, as VulnCheck expects.
- What leaked before the planned disclosure, and how.
Questions people ask
What is Java4Shell?
Java4Shell is the name Calif gave to CVE-2026-114938, a memory corruption flaw in how OpenJDK decodes Base64 data. Calif says it can leak memory or lead to remote code execution in applications that run on affected Java versions.
Is Java4Shell being exploited in the wild?
Infosecurity Magazine reported that no attacks had been seen as of 18 November, and Help Net Security reported no evidence of successful attacks. VulnCheck does not expect widespread exploitation, though it expects narrow paths to code execution in specific applications.
Which systems are vulnerable to CVE-2026-114938?
Calif says OpenJDK from 17.0.6 onward and distributions based on it are affected, and Infosecurity Magazine reported OpenJDK 17, 21, 25 and 27. Exploitation needs an x86-64 chip with AVX-512 VBMI. ARM64 systems are not vulnerable.
How severe is Java4Shell?
Calif gave it a CVSS 4.0 score of 9.2. Oracle gave it 8.1 on CVSS 3.1 and cited high attack complexity. The two scores use different versions of the scoring system, so they are not directly comparable.
Sources
- SecurityWeek: Java4Shell Vulnerability Puts Many Enterprise Applications at Risk of Takeover
- The Hacker News: Java4Shell Flaw (CVE-2026-114938) Enables Pre-Auth RCE in Jira, Jenkins, Keycloak and More
- Infosecurity Magazine: Java4Shell Vulnerability Allows Remote Attacks on Major Java Apps
- Help Net Security: Java4Shell OpenJDK flaw: Exploitation is tricky and hinges on specific CPUs (CVE-2026-114938)
- Security Affairs: 65 Characters to RCE: Inside the Java4Shell Vulnerability