A cyber-security start-up called Calif has used artificial intelligence to find a critical security flaw in Java — the language underneath an enormous amount of the world's banking, aviation and government software.
The bug sits in Java's memory functions. It has been named Java4Shell, and it allowed an attacker to remotely execute unwanted commands against Java's memory system.
Why this one matters more than most
Java is the world's third largest programming language. Around 90% of Fortune 500 companies use it — Apple and Amazon among them — along with a great many banks, hospitals and government agencies.
First released in 1995, it is what Calif called the "beating heart of the modern economy". Almost any time you transfer a file, log into a system with a password or move money online, something in that chain is likely running Java.
The particular cruelty of this flaw is where it sits. One of Java's main selling points is that it manages and protects memory on the programmer's behalf, precisely so that the programmer cannot make the classic mistakes.
Professor Toby Walsh, one of the world's leading AI researchers, told the BBC that this is exactly why the bug is serious. Java is meant to make the IT world more secure by handling memory for you. "This is why this is such an unlikely and serious flaw," he said.
Overloading a system's memory is one of the oldest and most reliable ways for an attacker to crash or hijack it. A memory flaw in the language built to prevent memory flaws is a bad category of problem.

The attacker was a model the public cannot use
The "adversary" that found Java4Shell was OpenAI's GPT-6 Cyber — a model that has not been released publicly.
That detail cuts both ways. A tool unavailable to the public found the flaw before attackers did, which is the whole argument for this kind of research. But the existence of the flaw is now public, and the fear among researchers is that attackers will go looking for it if it is not patched everywhere first.
Oracle shipped a patch without admitting the bug
Oracle, which owns Java, has not acknowledged Java4Shell. It did release a "critical patch update" on Tuesday, which it said contained security vulnerability fixes.
Whether any platform or company has actually been exploited is still unclear.
Cyber-security expert Dr Pedram Hayati said the flaw could compromise "critical financial systems, customers and organisation's data". In Australia alone, he added, almost every bank uses Java to manage critical transaction data.
The same company found the PHP flaw last month
Calif is the firm that identified the PHP vulnerability in October — the one we covered here as PHP2Shell, which left an estimated 30 million websites exposed, according to Alan Woodward, a professor at the Surrey Centre for Cyber Security.
Two critical flaws, in two foundational languages, found by one small company using AI, inside a month. That pattern is becoming the story in its own right.
Earlier in November, OpenAI president Greg Brockman said his company was moving 25% of its staff onto cyber-security work.
The timing is doing a lot of work here
This lands in the middle of a run of outages: a large Microsoft failure, a network fault that grounded US flights, full outages across several Apple services after an internal update, and customers of JP Morgan Chase, Wells Fargo and Bank of America all reporting trouble in the same week.
The US National Security Agency's internal communications also went down on Tuesday, which it attributed to "scheduled routine maintenance". It did not say whether the Java bug was involved.
US Secretary of State Marco Rubio has said there is "zero evidence" the incidents are connected, and nothing so far links any of them to Java4Shell. It is worth keeping those two things separate: a serious flaw was disclosed, and a lot of things broke in the same week. Neither fact is evidence for the other.
What to actually do
If you run anything on Java, apply Oracle's November critical patch update. That is the whole instruction, and it is the same one as October's PHP flaw: the fix exists, and the risk lives in the gap between release and installation.




