November 19, 2026, (Inside AI) — Security firm Calif has used artificial intelligence to uncover a critical vulnerability in Java that had gone unnoticed for 12 years. The flaw, tracked as CVE-2026-114938 and nicknamed Java4Shell, allows an unauthenticated attacker to remotely execute code on any system running affected Java versions. Oracle shipped a fix on November 17 as part of a broader update covering 652 security patches. Calif rated the flaw 9.2 out of 10 on the CVSS 4.0 scale. Anything above 9 counts as critical.
The discovery matters because Java runs nearly everywhere. Banks, hospitals, governments, airlines, and telecom operators depend on it for critical operations. Any time a user transfers money online, logs into a system, or receives a file, the platform likely relies on Java. A remotely exploitable flaw in the runtime is a potential skeleton key for attackers.
The bug sits in OpenJDK's Base64 decoder, specifically in hand-written assembly code that uses AVX-512 hardware acceleration. Datadog Security Labs describes it as an out-of-bounds write, meaning the code puts data where it does not belong. That can crash a program or, at worst, give an attacker full control over the system. The flaw affects OpenJDK versions 17, 21, 25, and 27, plus Oracle's GraalVM.
Researcher Quang Luong at Calif found the vulnerability using OpenAI's Codex in approximately one hour. The total cost came to under $200 in AI tokens. Luong then built the first working exploit in about two more hours. Calif published the technical details but withheld application-specific exploit chains to give organizations time to patch.
As explained by the Calif website:
"Today, we are disclosing Java4Shell (CVE-2026-114938), a memory-corruption vulnerability in OpenJDK's Base64 decoder. OpenJDK is the most popular Java runtime. Under CVSS 4.0, this vulnerability scores 9.2 out of 10 (CRITICAL)." (Calif, official website)
Oracle's November 17 Critical Security Patch Update addressed 11 Java SE vulnerabilities in total. All 11 can be exploited remotely without authentication. The company advised customers running Java 8, 11, 17, 21, 25, and 27 to upgrade immediately to patched replacement releases.
As of publication, no confirmed exploitation of Java4Shell exists in the wild. CISA has not added it to its Known Exploited Vulnerabilities catalog. However, cybersecurity experts warn that the window between patch release and active exploitation shrinks every year, especially when technical details become public.
The discovery follows Calif's earlier find of PHP2Shell, two critical vulnerabilities in the PHP programming language disclosed in October. Those flaws affected an estimated 30 million websites and went undetected for over a decade as well. Professor Toby Walsh noted that AI now exposes flaws "much more quickly," including ones so obscure that humans alone would not have found them.
Walsh called the discovery "such an unlikely and serious flaw" because Java's core design promise is managing memory safely.
AI Reshapes Vulnerability Discovery Economics
The Java4Shell find is not just a technical curiosity. It signals a shift in how vulnerabilities are found. A single researcher with an AI assistant located a critical flaw in a widely used runtime in about an hour for less than the cost of a dinner for two. That economics changes the calculus for both attackers and defenders. Previously, finding such a bug required deep expertise and weeks of manual code review. Now, the barrier to entry has collapsed.
Calif frames this as a new workflow. As per Calif:
"The same centaur hacker that discovered PHP2Shell and Java4Shell can comb a large codebase and flag dead code, bloated features, and needless optimizations. We call this vibe-uncoding: identify the parts a system actually needs, then remove everything else to reduce its attack surface." (Calif, official website)
The term "centaur hacker" refers to a human working in tandem with an AI model. The approach suggests that AI can be used not only to find bugs but also to simplify code and reduce attack surfaces. That could be a more durable defense than patching alone.
OpenAI president Greg Brockman announced earlier in November that his company shifted 25% of its staff to focus on cybersecurity demands. The Java4Shell discovery demonstrates why. AI tools accelerate both offense and defense, and the organizations that patch fastest will determine which side benefits more.
For now, the immediate priority is patching. Organizations running affected Java versions should upgrade to the latest releases. The flaw is critical, and the technical details are public. The clock is ticking.
Tobias is less interested in what a robot or a prompt can do today than in what it means for tomorrow. He writes about prompt engineering and robotics for readers who want context, not just specs.