Calif says it discovered a critical Java flaw with AI’s help
🎧 Listen to this story
Cybersecurity startup Calif identified, with help from GPT-6 Cyber, OpenAI’s model, and the Codex agent, a critical vulnerability in Java called Java4Shell (CVE-2026-114938). The flaw is in OpenJDK’s Base64 decoder and could cause crashes, expose information or enable remote code execution in certain installations. Calif demonstrated exploits in Keycloak, Atlassian Jira, Confluence, Jenkins and JFrog Artifactory, but the vulnerability affects only specific OpenJDK versions running on x86-64 processors compatible with AVX-512 VBMI; ARM64 systems are not considered affected.
Calif says researcher Quang Luong took about an hour to find the problem and two more hours to create the first working exploit, with AI costs below US$200. Oracle published a security update on November 17, rated the flaw as high severity, and the U.S. Cybersecurity and Infrastructure Security Agency (CISA) recommended installing it immediately. Authorities in Germany and the Netherlands said there were no known records of successful attacks as of November 18.
Why it matters · editorial interpretation
For teams using affected OpenJDK versions and configurations, the practical priority is to assess exposure and apply the update recommended by CISA. Calif’s account also suggests that coding agents may reduce the time needed to investigate and exploit vulnerabilities, although that conclusion comes from the company itself.
Sources
SecurityCodeAgents