Skip to content

AI

Calif used AI to discover critical Java flaw Java4Shell

Oracle released patched Java builds and CISA urged immediate updates the same day Calif disclosed the memory-corruption bug.

https://breach-signal.com/artificial-intelligence/calif-ai-discovers-java4shell-openjdk-vulnerability/

Share

Key facts

  • Calif, a cybersecurity startup, used artificial intelligence to discover a critical memory flaw in Java named Java4Shell and tracked as CVE-2026-114938 on Tuesday, 2026-11-17.
  • The flaw is a memory-corruption bug in Java's Base64 decoder in OpenJDK that lets an unauthenticated attacker corrupt memory and run unauthorised commands.
  • Oracle shipped patched Java builds on Tuesday afternoon and released a critical patch update; CISA published a notice late Tuesday urging organizations to apply the update immediately.
  • Java runs on around 90 percent of Fortune 500 companies and nearly all top companies including Apple and Amazon, and underpins banking, national security systems, aviation and telecommunications.

Calif used artificial intelligence to discover a critical memory flaw in Java named Java4Shell and tracked as CVE-2026-114938 on Tuesday, 2026-11-17.

The flaw sits in Java’s Base64 decoder in OpenJDK, a component that turns computer data into text format. Calif described it as a memory-corruption flaw that lets an intruder corrupt memory and run unauthorized commands. The buggy code miscalculates a length and writes 64 bytes into a buffer meant to hold as few as 50, damaging adjacent memory.

The bug activates when it sees 64 ‘A’ characters followed by one equal sign, affecting chips with AVX-512 only while leaving ARM64 systems untouched. The flaw lives in handwritten processor instructions that OpenJDK added to speed up Base64 decoding. Even after the program rejects the bad input, the surrounding memory stays damaged, and that corruption typically crashes the program by itself.

Calif showed that carefully arranging surrounding memory, a technique known as heap grooming, lets an attacker convert that corruption into control of the server. The startup also said an unauthenticated attacker may corrupt server memory, retrieve nearby data similar to the 2014 Heartbleed bug, and convert that access into complete control. Calif considers the flaw critical and rated it 9.2 on a 10-point CVSS scale. OpenJDK issued a security advisory on Tuesday, 2026-11-17.

Oracle put out patched Java builds on Tuesday afternoon and released a critical patch update. Oracle has not acknowledged the Java bug, though it issued a critical patch update on Tuesday with security vulnerability fixes covering 652 flaws across its product line. CISA published a notice late Tuesday urging organizations to apply the update immediately.

Calif chief executive Thai Duong said exploiting the vulnerability is difficult, giving organizations time to patch, though unpatched systems may carry risk for years. Gal Elbaz, co-founder and CTO of Oligo Security, said early exploit work took just a few hours, the technical barriers are surmountable, and organizations should treat this as a serious exposure.

Reports disagree on the AI model involved: the BBC and CNBCTV18 name OpenAI’s GPT-6 Cyber, while CyberScoop names OpenAI’s Codex. Calif said researcher Quang Luong found the flaw in roughly one hour while auditing OpenJDK with OpenAI’s Codex, spending less than $200 on computing tokens, and AI agents produced most exploit chains nearly on their own. The BBC described GPT-6 Cyber as released but not yet available to the public.

Calif showed unauthenticated code execution in Keycloak, Jira, Confluence, Jenkins and PingFederate, including a working break-in against Keycloak. It also demonstrated authenticated execution in Kafka, Solr and Elasticsearch. samczsun showed that connecting to a malicious server can compromise Java programs.

Java runs on around 90 percent of Fortune 500 companies and nearly all top companies including Apple and Amazon, and underpins banking, national security systems, aviation and telecommunications. Calif describes Java as the beating heart of the modern economy; the language was released in 1995. Professor Toby Walsh said Java primarily manages and protects memory for users. ‘this is such an unlikely and serious flaw,’ Walsh said.

People transferring information online, receiving files from users, or logging into IT systems with passwords are likely using a Java program. Dr Pedram Hayati said the serious flaw could put critical financial systems, customers and organizational data at risk. He also said nearly every bank in Australia uses Java to manage critical transaction data.

Calif worked with government agencies in the Five Eyes alliance and with Google and Microsoft on the response. Google said it rolled out mitigations and patches across its services after Calif notified it. Eduardo Vela Nava, an information security engineer at Google, said ‘Once notified, we swiftly rolled out mitigations and patches across Google services.’

Google created detection to confirm no exploitation occurred and gave customers guidance. The company moved quickly after Calif provided details. The New York Times verified the finding with its own testing in under an hour after Calif shared vulnerability details.

No exploitation in the wild had been reported and no related CVE was on CISA’s Known Exploited Vulnerability list when CyberScoop published, according to CyberScoop. The BBC reported that it was still unknown whether any platforms or companies had been affected by the latest Java incident.

An October 2026 Oracle urgent notice covered a different Java flaw, CVE-2026-91403, scored 9.1 on CVSS 3.1, exploitable over a network without login, reaching Java SE releases 8u511 / 11.0.33 / 17.0.21 / 21.0.13 / 25.0.5 / 27.0.1 plus GraalVM. After a patch ships, unpatched systems enter an N-day attack window, and applying the update right away is the best defense; Java and Node.js entered that window once patches were available.

Calif disclosed a PHP vulnerability, php2shell, last month. An estimated 70 percent of websites and apps use PHP. Professor Alan Woodward of the Surrey Centre for Cyber Security said the PHP flaw affected some 30 million websites, and Toby Walsh said the flaws remained unnoticed for at least a decade, unseen for 12 years.

On October 20 and 21, 2026, PHP emergency fixes were released for CVE-2026-109618 and CVE-2026-109619, both rated 9.2. Attack code was circulating and some firms reported attacks while the IPA had not confirmed any as of October 21.

Websites such as Wikipedia use PHP to collect information and present it in a readable form on the web. Website owners were urged to check whether the PHP flaw affected their sites and, if so, to install updates or consider going offline.

This discovery joins a rapidly expanding list of flaws surfaced by AI, with researchers increasingly relying on advanced AI models to uncover issues that experienced teams might take much longer to find. Researchers told CyberScoop that real-world exploitation is likely because attackers can quickly leverage artificial intelligence.

According to Toby Walsh, AI is uncovering additional vulnerabilities at a much faster pace, including some too obscure for humans to find without assistance. OpenAI president Greg Brockman said in November 2026 that the company would move 25 percent of staff to cyber-security work.

The Java4Shell disclosure on Tuesday came as several serious technology problems emerged, including a major Microsoft outage. Apple said multiple services went down completely on Tuesday because of an internal update, and all flights in the US were grounded by a network failure. Earlier in the week, thousands of customers at major American banks including Wells Fargo, Bank of America and JP Morgan Chase reported trouble accessing services.

NSA internal communications went offline on Tuesday, 2026-11-17, because of scheduled routine maintenance, and the agency did not specify whether it was linked to the Java bug. Secretary of State Marco Rubio said there was zero evidence the recent incidents were related.

These events showed that almost every critical sector depends on aging online infrastructure, including national security systems, aviation and telecommunications. Cybersecurity experts said flooding a system’s memory is one of the most common ways hackers exploit or crash a system.

Was this helpful?

Join the discussion

Please read our Comment Policy before commenting. First-time comments are held for moderation. By posting a comment you agree to our Terms of Use and Privacy Policy, which explain how your name, email address and comment are used.

Your email address will not be published. Required fields are marked *