AI finds critical Java4Shell flaw; CISA and South Korea tell users to patch now

AI finds critical Java4Shell flaw; CISA and South Korea tell users to patch now

A critical Java security flaw called Java4Shell, surfaced with AI assistance, has pushed the US and South Korea to issue urgent advisories telling users to install security updates. The vulnerability is tracked as CVE-2026-114938 and affects Java 17 through the OpenJDK implementation.

Security start-up Calif says it used OpenAI’s GPT-6 Cyber model to find the bug. The same start-up had earlier used AI to uncover a major PHP flaw, and reports now describe Java as the world’s third-largest programming language and a platform used by 90% of Fortune 500 companies.

What the flaw enables

The flaw is a remote code execution bug that can be triggered before authentication. Calif says a 65-character Base64 string can corrupt the Java heap on AVX-512 servers. The company reports pre-authentication RCE in Jira, Jenkins and Keycloak, three applications that many enterprises run on Java.

The impact, if exploited, is not limited to crashing a service. The reports say the flaw could expose sensitive data and allow an attacker to run code remotely on affected systems. That places servers, developer pipelines and identity-management deployments inside the same exposure zone.

Oracle’s patch and the advisories

Oracle released a patch for the Java security flaw and issued 11 Java SE security updates on the same day. CISA urged users to patch Java4Shell immediately because it can allow remote code execution. Reuters reported that US and South Korean authorities issued cybersecurity advisories on Nov. 18 asking Java users to install updates urgently.

The advisory push matters because the affected codebase is not a niche component. Java underpins key government agencies and major businesses, and the reports say millions of systems may be at risk if administrators delay.

How the flaw was found

Calif’s finding differs from a routine code review because of the role played by generative AI. The company says OpenAI’s GPT-6 Cyber model helped expose the vulnerability in Java. It is not the first time this start-up has made that claim: the same team previously used AI to find a flawed component in PHP.

The technique also accelerated discovery of a flaw that has been described as having existed for years. One report calls it a decade-old security flaw in Java, although the advisories and patch notifications are current.

Where the exposure lies

The risk concentrates on servers running Java 17. The 65-character Base64 trigger and the heap-corruption bug on AVX-512 hardware point to a highly specific attack path rather than a broad client-side issue. Applications named in the finding include Jira for issue tracking, Jenkins for continuous integration, and Keycloak for identity and access management.

For enterprises running those tools on unpatched OpenJDK builds, the immediate task is to apply Oracle’s Java SE fixes and verify the version across hosts. The advisories from Washington and Seoul make clear that the patch window is not a routine maintenance cycle.

Sources

About The Author