News
Java4Shell: Critical OpenJDK Flaw Found With Codex for Under $200, Calif Says

Researchers at Calif Research have disclosed Java4Shell (CVE-2026-114938), a flaw in OpenJDK's Base64 decoder that can lead to remote code execution. According to the team, a researcher using OpenAI's Codex found it in about an hour, and Oracle shipped patches on November 17.
Calif Research disclosed a critical OpenJDK flaw on November 17, dubbed Java4Shell and tracked as CVE-2026-114938. According to the team, researcher Quang Luong found it in about an hour using OpenAI's Codex coding assistant, with token costs of under $200. Oracle shipped patches the same day.
How the bug works
The flaw sits in hand-written code that speeds up Base64 decoding on processors with AVX-512 instructions. A malformed string causes a write past the end of the output buffer and corrupts adjacent data on the heap. According to CyberScoop, 64 "A" characters and a single equal sign are enough to trigger it. The input is rejected afterward, but the memory is already damaged.
Memory corruption on its own usually just crashes the program. With careful heap grooming, however, Calif says it can allow an attacker to read neighboring data in a manner similar to the 2014 Heartbleed bug, and to take over a server remotely without authentication.
Who is affected
According to CyberScoop, the vulnerable optimization runs only on processors that support the AVX-512 instruction set, and ARM64 systems are not affected. The bug lives in the Base64 decoder in OpenJDK, the Java runtime that, by Calif's count, runs at 90% of Fortune 500 companies.
Calif reported unauthenticated remote code execution in popular products including Keycloak, Jira, Confluence, Jenkins and PingFederate, and demonstrated a working break-in against Keycloak. CyberScoop adds that authenticated code execution was shown in Kafka, Solr and Elasticsearch.
According to Calif, there is no universal exploit, and each target requires tailored steps. The team is withholding product-specific exploit details so users have time to upgrade.
That doesn’t mean every Java application is vulnerable or that exploitation will be straightforward. - Gal Elbaz, Oligo Security
The role of AI
It is Calif's second high-profile AI-assisted find in recent weeks, after the php2shell flaw in October. According to Calif, AI agents built most of the team's exploit chains nearly on their own. Calif calls this the era of the "centaur hacker, part human, part AI."
The time and cost figures in the sources come from Calif itself. CyberScoop reports that the researcher used OpenAI's Codex.
What to do
CISA published a notice late Tuesday calling for the patch to be installed. As of CyberScoop's publication, no exploitation in the wild had been reported, and none of the related CVEs had been added to the KEV list. Researchers nonetheless rate the likelihood of real-world attacks as high.
Oracle shipped patched Java builds on November 17, and OpenJDK published a security advisory. Users should update their Java runtime.
Gal Elbaz of Oligo Security says organizations should treat the flaw as a serious exposure and apply compensating controls that block exploitation while patches are rolled out.
