Ana içeriğe geç

Researchers Used AI to Find a Critical Java Flaw

Calif researchers used OpenAI’s GPT-6 Cyber model to help find Java4Shell, a Java bug that can lead to remote code execution, and Oracle has released a patch that CISA urges organizations to apply immediately.
Dario Radečić
Dario Radečić

Ücretsiz Hesabınızı Oluşturun

Alıştırma and program your progress.

Google ile devam edinDevam ederek, Kullanım Koşullarımızı ve Gizlilik Politikamızı kabul etmiş ve verilerinizin ABD'de saklandığını onaylamış olursunuz.

A serious flaw was found in the software that about 90 percent of Fortune 500 companies use.

On November 17, 2026, the start-up Calif announced a major bug in Java, found with OpenAI's GPT-6 Cyber model. Java runs almost everywhere - from banks, hospitals, and airlines to government agencies. Oracle released a patch the same day, and CISA urged organizations to patch immediately.

Calif named the bug Java4Shell, and it's the second major flaw the team found with OpenAI models in the past couple of weeks. AI now helps small teams do vulnerability research that once took far more people and time, and that leaves you less time to patch.

In this article, I'll walk you through what Java4Shell is, how GPT-6 Cyber helped find it, how Oracle and CISA responded, and what you should do about it.

But what is Java anyway? Learn the language fundamentals with our beginner-friendly documentation.

What Is the Java4Shell Vulnerability?

Java4Shell is the name Calif researchers gave to a newly found bug in Java's Base64 decoder.

Base64 is a common method that converts binary data into plain text made of letters, numbers, and a few punctuation marks. Applications use it to send binary data through systems that only accept text.

On some computers, Java's decoder gets a bug when it reads a specific kind of input. The bug makes Java tamper with a section of memory it shouldn't access, which typically crashes the program.

But that crash isn't the worst outcome.

The heap is the part of memory where a program stores its data while it runs. By arranging that memory in a careful layout, a technique called heap grooming, Calif researchers turned the bug into a way to corrupt data on purpose. They showed that this can give an attacker full unauthorized access to the underlying system.

In plain English, that's the path to remote code execution, where an attacker runs their own code on your system from afar. The New York Times independently verified the attack after Calif shared proof-of-concept details, and the test took less than an hour.

Still, Java4Shell is a bug in Java's implementation of the decoder, not a flaw in the Java language itself.

How GPT-6 Cyber Helped Discover the Java Bug

Calif researchers used OpenAI's recently released GPT-6 Cyber model during their Java research.

The underlying bug in the Base64 decoder came from Quang Luong, a security researcher at Calif. Here's how the work was divided:

  • Calif researchers: Found the bug, directed the investigation, assessed the findings, and demonstrated the security impact
  • GPT-6 Cyber: Helped the team investigate the flaw and show what it could do

The reporting doesn't explain which steps the model handled. What it does say is that the most alarming AI-assisted discoveries still need human expertise.

That matters because of speed. Small teams of veteran security researchers can use top AI models to build, in as little as days or even hours, attacks that once took elite teams at agencies like the NSA months to do.

OpenAI's Fouad Matin, who leads cyber preparedness and safety at the company, said the Java research shows AI can assist expert researchers who look for serious flaws in widely used software.

Why the Java Vulnerability Is So Serious

Java4Shell matters because of where Java runs and what the bug can do. Let me explain.

Java's widespread use

Large organizations build much of their software on Java. That includes cloud platforms, identity management tools, and email services.

Java is everywhere

Thai Duong, CEO of Calif

A flaw in Java can reach any application that runs on an affected Java runtime, the software that executes Java programs.

Remote code execution

Remote code execution (RCE) means an attacker makes a system run code of their choice from a distance.

That's a bigger problem than a crash. A crash stops a service, but code execution lets an attacker act with the permissions of the application they took over. From there, they can read data, change it, or try to reach other systems on the same network. A flaw like this could easily give a hacker remote access to sensitive internal systems.

Authentication systems

Calif also showed that the bug can break into Keycloak, a widely used open-source system for authentication and log-in. Keycloak protects thousands of organizations, including CERN, Hewlett Packard, UnitedHealthcare, and NASA.

That makes it a high-value target. Every application behind it relies on it to check who's allowed in.

Long-term exposure

A patch fixes the bug only where someone installs it.

Duong warned that the bug could threaten systems that skip updates for the next five to 10 years. Java runtimes are often bundled inside other products and containers, so they're easy to miss. Organizations that delay updates stay exposed for as long as the old runtime stays in use.

But widespread use of Java doesn't mean every Java application is vulnerable. Standard Java 8 and 11 don't contain the affected decoder code. Exposure also needs an x86-64 processor with a feature called AVX-512 VBMI, plus data an attacker controls that reaches the decoder.

How Oracle and Security Agencies Responded

Oracle released a security patch on November 17, the same day Calif announced the bug. It's tracked as CVE-2026-114938.

Tenable's breakdown of Oracle's advisory lists these builds as affected, and earlier versions are likely affected too:

  • Java 17: 17.0.21.0.3
  • Java 21: 21.0.13.0.3
  • Java 25: 25.0.5.0.3
  • Java 27: 27.0.1.0.3
  • Java SE 8 performance build: 8u511b33-perf

The fixed builds are 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1, and 8u513 for the Java SE 8 performance build.

CISA pressed organizations to patch immediately.

Calif worked with every agency in Five Eyes, the intelligence-sharing alliance of the US, Canada, Australia, New Zealand, and Britain. It also coordinated with major vendors, including Google and Microsoft.

Eduardo Vela Nava, an information security engineer at Google, said the company rolled out mitigations and patches across its services. Google also built detection to check for exploitation and gave guidance to its customers.

Duong said the bug isn't easy to weaponize, which gives defenders more time to patch.

A couple of unrelated outages hit Microsoft products, flights, and banks around the same week. These different events seem to have different explanations, and those explanations each look benign so far, so we don't connect them to Java4Shell.

How AI Is Changing Cybersecurity Research

AI models now find flaws in code so fast that security professionals struggle to keep up.

A model can quickly read a lot of code, flag behavior that looks unusual, and help a researcher test an idea. That automates the repetitive parts of the work, so a team gets to a finding faster.

Java4Shell isn't Calif's first big discovery. In October 2026, the start-up found serious flaws in PHP with the help of OpenAI models. (PHP is a programming language used by about 70 percent of websites and apps.)

Government agencies, including the NSA, warned businesses to take their websites offline if they couldn't patch immediately. The disruptions proved less serious than some predicted, because hackers couldn't easily reverse-engineer the attack method.

These results change what a small team can achieve. Calif and other groups that use AI tools have upended what people considered possible for a small group of coders just a few months ago. Basically, AI has made it much easier to find weaknesses in systems. Then, the vulnerabilities can be severe and deep in a codebase, and researchers can make mistakes when fixing them, which can take websites offline.

What Organizations Should Do About Java4Shell

Here's what you should do, in order:

  1. Read the advisories: Review Oracle's security advisory for CVE-2026-114938 and CISA's alert
  2. Find your Java runtimes: List every runtime and application that uses Java, starting with Java 17 and later, including runtimes bundled inside other products and containers
  3. Check vendor advisories: Look at the advisories for any Java-based product you use
  4. Prioritize exposed systems: Start with internet-facing and sensitive systems, such as log-in and identity services
  5. Apply the updates: Install the fixed build for your Java version
  6. Review detection guidance: Check what your security vendors and Java distributor publish for this CVE
  7. Monitor for exploitation: Look for signs of attempted exploitation in your logs and network traffic

The exact fix depends on your Java version and how you deploy it. A server with its own JDK, a container image, and a product that bundles a runtime will each need a different update path.

Conclusion

GPT-6 Cyber didn't do this alone. Quang Luong found the bug, and Calif's researchers directed the work and demonstrated its impact. The model supported a human-led effort and didn't replace the expertise behind it.

But the same capabilities work for attackers. In November 2026, the FBI warned that hackers use new autonomous AI tools to target government agencies and businesses. Calif's CEO, Thai Duong, says it could be three months or three years before AI can build the most advanced attacks on its own, and that could leave defenders less time to patch.

The question is whether AI-assisted defense can keep pace with AI-assisted attacks. We'll have to wait and see.

Read our blog post if you want to learn more about AI in Cybersecurity from a researcher's perspective.

Dario Radečić

Yazar

Dario Radečić

Senior Data Scientist based in Croatia. Top Tech Writer with over 700 articles published, generating more than 10M views. Book Author of Machine Learning Automation with TPOT.

FAQs

What is Java4Shell?

Java4Shell is a bug in Java's Base64 decoder that Calif researchers named and reported. Under particular conditions, it corrupts memory and can lead to remote code execution. Oracle released a patch on November 17, and the bug is tracked as CVE-2026-114938.

How did GPT-6 Cyber help find the bug?

Calif researcher Quang Luong found the bug in the Base64 decoder. GPT-6 Cyber helped the team investigate the flaw and show what it could do. Researchers directed the work and assessed the findings, so the model supported a human-led effort.

Is Java4Shell dangerous?

It's serious because Java runs inside banks, hospitals, and government agencies, and the bug can lead to remote code execution. But not every Java application is vulnerable or exploitable. Calif's CEO, Thai Duong, says the bug isn't easy to weaponize, which gives defenders time to patch.

Which Java versions are affected by Java4Shell?

Oracle lists builds of Java 17, 21, 25, and 27 as affected, plus its Java SE 8 performance build. Standard Java 8 and 11 don't contain the affected decoder code. Exposure also needs an x86-64 processor with AVX-512 VBMI support. The fixed builds are 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1, and 8u513 for the performance build.

How do I check if my systems are affected?

List every Java runtime and application you run, including runtimes bundled inside other products and containers. Then compare the versions with the builds in Oracle's advisory and check your vendors' advisories. Patch internet-facing and sensitive systems first, and look for signs of attempted exploitation in your logs.

Konular

İlgili