Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

calif.io+1.calif.io.calif.io+2.Cybersecurity startup Calif disclosed a critical vulnerability in Java on Tuesday, Nov. 17. The company said it found the flaw with help from OpenAI's recently released GPT-6 Cyber model, [according to The New York Times](https://www.nytimes.com/2026/11/17/us/politics/ai-security-bug-java.html) The New York Times Company . Java is used by government agencies and by about 90 percent of Fortune 500 companies. The flaw, called Java4Shell and tracked as CVE-2026-114938, is a memory-corruption bug in the Base64 decoder of OpenJDK, the most popular Java runtime. It scores 9.2 out of 10 on the CVSS 4.0 severity scale, [Calif said](https://calif.io/java4shell/). Oracle released a patch at 1 p.m. Pacific time on Tuesday, and Calif urged "all Java users" to update immediately.nytimes.com+1
The flaw is in hand-written assembly code that OpenJDK uses to speed up Base64 decoding with AVX-512 instructions on x86-64 processors. Malformed input, such as 64 "A" characters followed by a single "=", makes the routine write past the end of its memory buffer, [Calif explained](https://calif.io/java4shell/). The company said it achieved remote code execution without login credentials in the default setups of Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate. It also achieved code execution after login in Apache Kafka, Solr and Elasticsearch. Affected versions include Java 17, 21, 25 and 27 running on servers with CPUs that support AVX-512 VBMI, [according to BreachHistory](https://breachhistory.com/blog/java4shell-cve-2026-114938-openjdk-base64-rce). [Semgrep](https://semgrep.dev/blog/2026/java4shell-openjdk-base64-vulnerability/) noted that starting the JVM with AVX-512 disabled is a temporary workaround, but it recommended patching even on hardware that is not affected.semgrep.dev+2
Calif researcher Quang Luong found the bug in about an hour and built a first working exploit in two more hours, [the company said](https://calif.io/java4shell/), at a total AI token cost of under $200. The company used a technique called "heap grooming" to arrange memory so that the corruption gave it full unauthorized access to the underlying system, [the Times reported](https://dnyuz.com/2026/11/17/a-i-helps-find-major-security-bug-in-java-a-software-platform-vital-to-the-internet). In recent days, Calif worked with agencies in the Five Eyes intelligence alliance and with companies including Google Alphabet Inc. and Microsoft , [the Times said](https://www.nytimes.com/2026/11/17/us/politics/ai-security-bug-java.html).calif.io+2
Java4Shell is Calif's second major AI-assisted discovery in a few weeks. Last month the company disclosed PHP2Shell, a pair of critical PHP flaws that it first found in August with OpenAI's GPT-5.6 Sol through OpenAI's Patch the Planet program, [Fortune reported](https://fortune.com/2026/10/21/ai-tools-discovered-the-php-vulnerability-that-could-take-down-70-of-the-internet-now-the-race-is-on-to-patch-it-before-ai-powered-hackers-attack/). Wiz also found those flaws independently, [OpenAI said](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities). The PHP disclosure led many companies to take their websites offline while they patched. Separately, [Forbes reported](https://www.forbes.com/sites/thomasbrewster/2026/11/17/openai-gpt-cyber-finds-dozens-of-vulnerabilities-in-apple-tech/) on Tuesday that Calif had used GPT-6 Cyber to find about 35 vulnerabilities in Apple products.openai.com+2
Calif described its approach as the work of a "centaur hacker," part human and part AI. "AI brings the horsepower. Our researchers bring the context," [the company wrote](https://calif.io/java4shell/).calif.io