Security teams at roughly 90% of Fortune 500 companies face patching after a startup used OpenAI’s GPT-6 Cyber to uncover a bug in Java itself, according to the New York Times on Tuesday.
The flaw, dubbed Java4Shell, sits in the Java routine that converts binary to text. The New York Times, which replicated the vulnerability independently, broke the story of the find by the security startup Calif.
Calif’s researchers, using GPT-6 Cyber that OpenAI released earlier this month, showed that a character string — 64 As and then an equals sign — triggers the bug in the Java routine that converts binary to text. Normally, that string just crashes a Java program. Under the right conditions, it can do far worse.
Thai Duong, the CEO of Calif, told the New York Times that vulnerabilities have a very long tail.
The trigger sequence is: “AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=”. What makes it dangerous is the pairing. By pairing Java4Shell with a method known as “heap grooming” or “heap feng shui” — first shown off back in 2007 — an attacker can eventually seize full control of the vulnerable machine.
Duong did acknowledge that translating the vulnerability into an attack would not be easy, which likely gives administrators time to perform the updates on a non-emergency basis.
Calif is having quite a month. Last month, using GPT-5.6 Sol, the startup publicized a PHP vulnerability it had found that likely affects millions of websites. This time, the target was what Calif researcher Bruce Dang called the “beating heart of the modern economy.”
Oracle, which owns the trademark on Java, has acknowledged the problem, and there is already a patch. The U.S. Cybersecurity & Infrastructure Security Agency went further, issuing an urgent update.
CISA’s guidance is: “Review your software inventory for JDKs and Java runtimes, including those bundled with third-party applications or included in container images and appliances,” CISA urges.
The Times puts the figure at “about” 90% of Fortune 500 firms running Java — a believable number, though its origin is not clear.
GPT-6 Cyber was released earlier this month. Calif used it to spot a consequential bug in Java, following a PHP vulnerability found last month using GPT-5.6 Sol.
By pairing Java4Shell with a method known as “heap grooming” or “heap feng shui” — first shown off back in 2007 — an attacker can eventually seize full control of the vulnerable machine. Calif found this using GPT-6 Cyber.
- Per CISA, check your software inventory for JDKs and Java runtimes, including ones that ship with third-party applications or come inside container images and appliances.
- Install the patch Oracle has already issued; Duong conceded that turning the flaw into a working attack would not be simple, which probably buys administrators time to roll out the updates without treating them as an emergency.
- Calif used GPT-6 Cyber, released earlier this month, to spot the bug.
Oracle, which owns the trademark on Java, has acknowledged the problem, and there is already a patch.
In this case, the startup uncovered a flaw in what Bruce Dang, its researcher, described as the beating heart of the modern economy.