Java4Shell: Researchers Used AI to Find a Critical Java Flaw That Lets Attackers Run Code Remotely; Oracle Has Patched It

Oracle patched Java4Shell, a Java flaw that lets attackers run code remotely, found with AI by Calif. Who is exposed and what to update.

Share this article

Featured photo: Server racks in a data center (file photo; not a system affected by Java4Shell). Photo: Victor Grigas, Wikimedia Commons, CC BY-SA 3.0.
8.1
Oracle severity score (CVSS 3.1)
9.2
Calif severity score (CVSS 4.0)
11
Java SE fixes in Oracle’s November update
~1 hour
Time to find the bug with AI help, per Calif

A security start-up called Calif used artificial intelligence to find a critical flaw in Java, the programming language and runtime used by banks, hospitals, governments and most large companies.[1] The flaw, which Calif calls Java4Shell, can let an attacker remotely run commands on an affected system.[1] Oracle, which owns Java, fixed it in a Critical Security Patch Update released on Tuesday, Nov. 17.[2][3]

Related: NSA Network Outage Feeds Hack Fears, but Officials Say It Was Maintenance, Not a Cyberattack.

Advertisement

What the flaw is

Java4Shell is tracked as CVE-2026-114938.[2] It is a memory-corruption bug in the OpenJDK code that decodes Base64 data.[3] The bug sits in hand-written assembly code that speeds up decoding on newer x86-64 processors.[3] Malformed input can make the routine write past the end of its memory buffer before Java rejects that input.[3]

Only x86-64 machines with a processor feature called AVX-512 VBMI are exposed.[3] That covers Intel Ice Lake or later and AMD Zen 4 or later chips; ARM machines are not affected.[3] Early reports name OpenJDK 17.0.6 and later as affected.[4]

Oracle scores the flaw 8.1 out of 10 under the CVSS 3.1 system.[2] Calif rates it 9.2 under the newer CVSS 4.0 system.[3] Both agree an attacker does not need to log in to exploit it.[3]

Which products are at risk

Calif said its testing achieved remote code execution without logging in against Jira, Confluence, Jenkins and Keycloak in their default settings.[3] Calif also said that “there is no universal exploit,” because each application uses Base64 differently.[3]

Oracle lists affected versions of Oracle Java SE, Oracle GraalVM for JDK and Oracle GraalVM Enterprise Edition.[2] OpenJDK builds from other vendors, Linux distribution packages and Java runtimes bundled inside applications or container images can also be affected.[4]

No exploitation in the wild had been reported, and Calif has not released its exploits.[3] Cloudflare said it detected no signs of exploits on its network.[4] It is unclear whether any platforms or companies have been affected.[1]

How AI found it

Calif researcher Quang Luong found the bug in about an hour with AI help and built a working exploit in two more hours, for under $200 in AI computing costs.[3] Oracle’s advisory credits Luong of Calif.io and a researcher named Albert Yang with reporting the flaw.[2]

Reports differ on which AI model was used. The BBC reported that the model was OpenAI’s GPT-6 Cyber, which is not yet available to the public.[1] Daily CyberSecurity reported that Calif did not name the AI model it used.[3]

Advertisement

Calif earlier found a critical flaw in the PHP web programming language.[1] “It’s been hiding in plain sight for 12 years,” Toby Walsh, an AI researcher and computer scientist, told the BBC of the Java bug.[1]

What to do

Oracle’s November update contains 655 security patches, including 11 for Java SE.[2] Oracle said all 11 Java SE flaws may be exploited over a network without user credentials.[2]

Cloudflare advised updating every Java runtime to Oracle’s Nov. 17 build or a vendor’s fixed release, starting with internet-facing services.[4] Each updated service must be restarted, because a running Java process keeps the old code.[4] Cloudflare said it began switching off the affected Base64 routine across its Java workloads about an hour after the patches were published.[4] Some researchers have suggested a similar temporary setting, which neither Oracle nor OpenJDK has confirmed.[3]

For more coverage, see The Boston Report’s cybersecurity coverage and its technology and artificial intelligence sections.

Sources

  1. BBC News: Critical flaw found in vital internet language Java through AI
  2. Oracle: Oracle Critical Security Patch Update Advisory – November 2026
  3. Daily CyberSecurity: Java4Shell (CVE-2026-114938): Java SE Remote Code Execution Explained
  4. Cloudflare Radar: Java4Shell: remote code execution in OpenJDK

About the reporter: Theo Lin is an AI reporter covering technology, AI and science for The Boston Report. This article was produced with AI under our editorial policy. Read our AI and media disclosure.

Read next in Artificial Intelligence

Get The Boston Report Briefing

Our free weekly email with the most important stories from Boston and the world. Unsubscribe anytime.

Free. We never sell your address. Privacy policy

Get The Boston Report Briefing
Our free weekly email with the most important stories from Boston and the world. Unsubscribe anytime.
Sign up