AI Finds Java4Shell Bug Hidden for 12 Years
By @sharedot · · 8 pages
- Programming
- Cybersecurity
- Java
- AI
Startup Calif's GPT-6 Cyber AI found Java4Shell, a critical Java flaw undetected for 12 years.
A Decade-Old Flaw Found in One Hour
Cybersecurity firm Calif discovered a critical memory corruption vulnerability in Java, tracked as CVE-2026-114938 and dubbed Java4Shell, that the BBC reports had existed for more than a decade.
According to The Record, researcher Quang Luong found the bug in roughly one hour using OpenAI's unreleased GPT-6 Cyber model, developed a working exploit in two more hours, and spent less than $200 in total compute.
Oracle credits Luong and Albert Yang of Calif with the discovery in its November patch release.
Why an AI-Found Bug Is a First Worth Watching
The surprising reversal here is not just the bug but who found it: a machine.
Professor Toby Walsh told the BBC that AI is now exposing flaws much more quickly, including ones so obscure that humans could not have discovered them alone, and that Java4Shell had been 'hiding in plain sight for 12 years.'
The Evidence: Memory Corruption in the JVM
Streamline Feed reports that Calif's Luong identified the flaw inside the OpenJDK HotSpot JVM's Base64 decoder: on x86-64 processors with AVX-512 vector instructions, malformed Base64 input can trigger a 62-to-63-byte out-of-bounds heap write, potentially enabling unauthenticated remote code execution when combined with heap grooming.
Oracle rated the bug 8.1 (High) under CVSS 3.1, while Calif rated it 9.2 (Critical) under CVSS 4.0. CISA issued a Tuesday-night notice warning the bug could crash the Java Virtual Machine and permit unauthenticated remote code execution.
The Stakes: Java Underpins the Modern Economy
The BBC reports Java runs at nearly all of the world's top companies, including Apple and Amazon, plus banks, hospitals and governments, and Calif says it powers about 90% of Fortune 500 companies.
According to The Record, CISA urged organizations to audit software inventories for affected Java installations, including copies bundled in third-party applications, prioritizing internet-facing systems.
Streamline Feed reports exposed deployments include Red Hat Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins, and PingFederate — with no confirmed exploitation in the wild as of 17 November.
Not Connected to This Week's Outages
The disclosure landed alongside a noisy week of disruptions — Microsoft cloud outages, an FAA network failure that grounded US flights, Apple's internal-update glitches, and the NSA's internal communications going offline during what it called scheduled routine maintenance.
The BBC reports US Secretary of State Marco Rubio said there was 'zero evidence' the incidents were related, and Streamline Feed notes officials also rejected linking the Java bug to the outages. Confusingly, the NSA did not specify whether its outage was connected to the identified Java flaw.
What Comes Next: Patch, Mitigate, Adapt
Oracle released a critical patch update on Tuesday containing fixes for the flaw, which Streamline Feed says shipped alongside 651 other security fixes across OpenJDK 17.0.6 and later, 21, 25, and 27.
Teams that cannot immediately restart production can apply temporary JVM flags — -XX:-UseBASE64Intrinsics or -XX:UseAVX=2 — to disable the vulnerable code path at some decoding throughput cost, per Streamline Feed.
The Record reports Calif is withholding weaponized, application-specific exploits until everyone has had a chance to upgrade.