An artificial intelligence model has discovered a devastating vulnerability in the widely used Java programming language in just three hours, triggering an emergency alert from US cyber authorities.
Cybersecurity firm Calif reported yesterday that researchers used OpenAI's GPT-6 Cyber model to pinpoint the "Java4Shell" flaw (CVE-2026-114938) hidden deep within Java's underlying code.
The AI identified the design flaw in just one hour and spent two more hours automatically writing the exploit code. The total computing cost was under US$200 (HK$1,560).
Experts noted that such complex reverse-engineering previously required top-tier human hackers weeks or months to execute.
The vulnerability, rated a "critical" 9.2 out of 10, lies in a hardware acceleration feature of the Java Runtime used to process data on high-end processors like Intel Xeon and AMD EPYC.
Hackers can exploit the flaw by sending a specially crafted string to trigger an out-of-bounds memory write. This can crash servers or, in extreme cases, allow attackers to bypass authentication and execute malicious commands remotely.
Major enterprise software, including Atlassian Jira, Jenkins, and Keycloak, are affected.
Oracle and OpenJDK have released emergency updates. IT departments are strongly urged to upgrade their Java environments immediately, or temporarily disable the acceleration feature using a specific startup command if an immediate restart is not possible.
Security experts warned that AI has slashed the cost of discovering deep hardware and memory vulnerabilities, urging companies to establish automated defenses to counter the escalating speed of cyber threats.
Staff reporter