Incident sheet · /security
- System
- Security & Risk
- Filed
- Read
- 4 min · 853 words
- Severity
- Routine
Java4Shell: AI found critical OpenJDK flaw in one hour, under $200
Calif found CVE-2026-114938 in OpenJDK's Base64 decoder in one hour with under $200 of AI tokens. The 9.2-severity flaw enables pre-auth RCE in Jira, Confluence and Jenkins.

Event log
- CVE-2026-114938 scores 9.2/10 under CVSS 4.0 in OpenJDK's x86-64 Base64 decoder.
- Discovery took one hour and under $200 in AI token costs; exploit development took two more hours.
- Pre-auth exploits demonstrated against Jira, Confluence, Jenkins, Artifactory and Keycloak.
- Oracle and OpenJDK shipped coordinated patches on Nov. 17; CISA urged immediate patching.
- Google and Anthropic are compressing disclosure timelines from 90 days to as little as 30.
A researcher needed one hour and less than $200 in AI token costs to find a critical memory-corruption flaw in OpenJDK, the most widely deployed Java runtime. Security firm Calif disclosed the vulnerability, dubbed Java4Shell and tracked as CVE-2026-114938, on Nov. 17 alongside coordinated patches from Oracle and the OpenJDK Vulnerability Group.
The bug sits in OpenJDK's Base64 decoder for x86-64 systems. Under version 4.0 of the Common Vulnerability Scoring System (CVSS, the industry severity scale), it rates 9.2 out of 10. Specially crafted input can bypass memory boundaries, letting attackers leak sensitive heap data in a manner comparable to Heartbleed, or execute arbitrary code through routine web requests and authentication flows.
Java ranks as the world's third most popular programming language. It powers 90% of Fortune 500 companies and spans banking, aviation, telecommunications and national security systems. While Java is designed to prevent memory corruption, its performance-critical components still rely on low-level machine code, and that is where the flaw lives.
How fast did AI find it?
Calif, which played a leading role in uncovering last month's PHP2Shell memory-safety vulnerabilities, attributed the discovery to AI-driven code auditing.
"It took our own Quang Luong an hour to discover the vulnerability, and two more hours to develop the first working exploit," Calif said in a technical blog post. "The total token cost was less than $200."
The economics are the story. Finding critical bugs used to demand weeks of expert manual review. Calif's result suggests large language models can now do the work at a marginal cost approaching zero.
What systems are exposed?
Calif demonstrated remote code execution against unpatched instances of widely deployed enterprise applications in default configurations. Pre-authentication exploits hit:
- Atlassian Jira and Confluence
- Jenkins
- JFrog Artifactory
- Keycloak
Post-authentication attacks targeted Apache Kafka Broker, Apache Solr and Elasticsearch.
The Swiss National Cyber Security Centre (NCSC) cautioned the list is not complete. "Other Java applications, including banking software and other industry-specific solutions, can also be affected. Operators must clarify their exposure with their software suppliers," the NCSC said.
Justin Moore, director of adversary operations at Arctic Wolf, outlined the blast radius. "If successfully exploited, the vulnerability could allow an attacker to operate with the same privileges as the affected application, potentially providing access to sensitive data, credentials, cloud resources, identity systems or software development infrastructure," he said.
Saeed Abbasi, manager at Qualys' threat research unit, warned code execution is not the only outcome. "A successful exploit chain can read Java heap memory before it runs any attacker code," he said. "That memory can hold private keys, tokens and personal data. CISA also warns that crafted input can crash the JVM, so availability is at risk too."
Researchers note the flaw's real-world exploitability is limited by strict hardware, software and application requirements. "The nature of this bug makes it slightly more difficult to monetize for cybercriminals, but this buys you days, not months," said Chris Fistonich, senior manager of Malwarebytes Threat Labs.
What does CISA recommend?
In a news alert, CISA strongly urged organizations to immediately patch Java4Shell and take four actions:
- Identify affected Java installations, including JDKs and runtimes bundled with third-party applications.
- Apply vendor patches as soon as available, prioritizing internet-facing systems.
- Review Oracle's security advisory for affected products and versions.
- Monitor for indicators of compromise: rogue runtimes, unexpected JVM crashes and suspicious Java processes.
"If immediate patching is not feasible, implement temporary compensating controls, such as restricting and validating Base64 input at the application boundary, enforcing input-size limits and limiting unauthenticated access," CISA advised.
The NCSC told end users to promptly install updates through official update functions, noting embedded Java components often update via software or device updates.
Why are disclosure timelines shrinking?
Java4Shell lands as AI reshapes attacker economics. Google recently warned that attackers now fully automate vulnerability research and exploitation with zero human intervention. Google and Anthropic have responded by cutting coordinated vulnerability-disclosure timelines from the traditional 90-day embargo to as little as 30 days.
"A vulnerability reported today presents a much higher immediate risk to users than one reported last year," said Natalie Silvanovich, security engineer at Google Project Zero. "Aspects of our policy that require vendor negotiation or agreement, including determining whether a bug is fixed and whether a grace period applies, are quickly becoming unmanageable."
Fistonich framed the shift for defenders. "The bottleneck has shifted," he said. "Finding critical bugs used to be the slow, expensive part. Now, with AI tools, it's the rapid part, and patching is where we're falling behind. This is very likely to be the new normal and not merely a one- or two-off event."
Calif argued the deeper fix is reducing what needs patching: bloated codebases carrying unused dependencies. The company issued a call for "vibe uncoding." "If you don't need something, there is little reason to keep it running in your systems," Calif said. "Do you need that hand-rolled assembly Base64 decoder to process a few hundred bytes? Probably not."
via techtarget.com (Source)
More from Rebecca Stone
Show full bio
Senior reporter covering money, funding and market shifts around enterprise information technology and the leaders who run at CIO Monitor.
199 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- Java4Shell Flaw Scores 9.2, Exposes Heap Memory Across Enterprise Estates
- Java4Shell Flaw Exposes Jenkins, Jira and Keycloak to Remote Code Execution
- Microsoft's October Patch Tuesday Ships 1,408 CVEs, One Actively Exploited
- SAP patches 35 flaws, including four rated maximum severity
- AI Squeezes the Window for Vulnerability Remediation