AI Found a Critical Java Flaw in an Hour. Now Everyone Must Patch
Security researchers found a critical flaw in Java with help from an AI coding tool, and the race to patch it is now on. Calif, the firm behind the discovery, says the bug can let attackers run code on servers without logging in.
The Flaw
The bug, tracked as CVE-2026-114938 and nicknamed Java4Shell, sits in the Base64 decoder of OpenJDK’s HotSpot Java Virtual Machine. A string of crafted input makes the decoder write past the end of its output array, corrupting nearby memory on the Java heap.
Calif rates the flaw 9.2 out of 10 under the CVSS 4.0 scale, squarely critical. Oracle scores it 8.1 under the older CVSS 3.1 scale, and the OpenJDK advisory carries the same figure. Both agree an attacker needs no credentials to try it.
The vulnerable code runs only on x86-64 processors with AVX-512 VBMI support, such as recent Intel Xeon and AMD Ryzen chips. It affects OpenJDK 17.0.6 and later, including Java 21, 25 and 27, and Oracle’s Java 8 Enterprise Performance Pack.
What It Can Do
Calif says it achieved remote code execution without logging in against default setups of Keycloak, Jira, Confluence, JFrog Artifactory, Jenkins and PingFederate. With a login, it reached the same result in Apache Kafka, Apache Solr, Elasticsearch, Gerrit and Spinnaker too.
The flaw can leak memory from the Java heap, a trick Calif calls JavaBleed. What comes out can include personal data, private keys and tokens, and the firm says a Java program opening a TLS connection to a malicious server can be compromised too.
Calif researcher Quang Luong found the bug in about an hour while auditing OpenJDK with OpenAI’s Codex assistant, then built a first working exploit in about two more hours. The whole exercise cost under 200 dollars in AI usage, according to the firm.
The Patch
Oracle released fixes on 17 November in its November Critical Security Patch Update, which covers 655 flaws across its products. The OpenJDK project published an advisory the same day, and Amazon has shipped fixed builds for its Corretto Java distribution.
The US Cybersecurity and Infrastructure Security Agency urged organisations to patch immediately, warning that crafted Base64 input could crash the Java Virtual Machine and allow unauthenticated remote code execution. India’s CERT-In and Canada’s Cyber Centre issued alerts of their own.
No attacks using the flaw had been reported as of Tuesday, and it is not on CISA’s list of known exploited vulnerabilities. But Calif published its write-up only after the flaw leaked early, so the clock for attackers started early.
Our Read
The headline is the speed. A flaw in one of the most widely used runtimes was found in an hour and exploited in three, for less than the price of a lunch. That is an advance for defence, and a warning about how cheap offence has become.
Calif did the right thing by holding back its product-specific exploits while users patch. But the write-up gives away the trigger, and scanners will be probing for unpatched Java by the weekend. The patch window here is measured in days, not months.
Why It Matters
Java underpins cloud platforms, identity systems, email and banking software, so a runtime flaw reaches further than a bug in any single product. The same researchers found PHP flaws last month, and the pattern is clear: AI-assisted auditing is finding the bugs faster than vendors can fix them.
Background
Calif has made a habit of AI-assisted finds, including the PHP2Shell flaws disclosed in October. The Java disclosure follows other AI-security incidents this month, and Oracle credits Calif researchers Quang Luong and Albert Yang with reporting the flaw. Other vendors are still shipping their fixed builds.
Source: The Indian Express explainer and BleepingComputer’s report.


0 Comments