Java vulnerability — AI-Powered ‘Centaur Hacker’ Discovers Critical Flaw

Java vulnerability
Reuters=Yonhap News

A high-risk security vulnerability has been identified in Java, a core programming language widely used across corporate and public information systems, allowing attackers to remotely take over servers without requiring a login. This critical flaw was discovered by a security firm in just one hour using artificial intelligence (AI) tools, heightening fears that AI-driven security threats are spreading rapidly across national infrastructure.

Oracle and the open-source Java (Open JDK) community distributed an emergency security patch for this vulnerability. Affected products include Oracle Java SE and GraalVM, a Java development and execution environment. South Korea’s Ministry of Science and ICT (MSIT) and the Korea Internet & Security Agency (KISA) also issued urgent inspection and security update recommendations. The ministry warned that some programs allow attackers to remotely execute arbitrary code without authentication, urging organizations to prioritize inspections of internet-exposed services, groupware, and Enterprise Resource Planning (ERP) systems.

The vulnerability was discovered by the US security firm Califf using OpenAI’s AI coding tools. The issue stems from the process of decoding data that Java previously converted into text for network transmission. To accelerate processing speed, a specialized central processing unit (CPU) instruction encountered a manipulated string and overwrote adjacent data beyond its designated memory boundary. Exploiting this vulnerability allows attackers to remotely execute malicious code without login credentials or extract sensitive information, such as personal data and encryption keys, left in server memory.

According to Califf, a single researcher used an AI coding tool to analyze core Java code and discover the vulnerability in just one hour. An actual exploit code was generated two hours later. Califf stated that the entire process cost less than $200 in AI usage fees and named the vulnerability ‘Java4Shell’. The firm added that today’s best hackers are neither purely human nor purely AI, but ‘Centaur hackers’—half human and half AI.

Java serves as a core programming language driving internal infrastructure for businesses and institutions. A representative from a domestic security firm noted that while small and medium-sized enterprises frequently used PHP, which previously suffered vulnerabilities, Java is primarily used by large corporations and public institutions, amplifying the potential impact. It is widely utilized as the foundational program for financial sectors such as banks, aviation, telecommunications, defense, and enterprise ERP and groupware systems. According to the Ministry of the Interior and Safety’s 2026 statistics report on government-wide EA-based public information resources, 71.3% of the 15,618 public information systems at the end of last year were built on Java, with central administrative agencies reaching 89.1%.

Califf evaluated the risk of this vulnerability at 9.2 out of 10 on the Common Vulnerability Scoring System (CVSS) 4.0 scale, classifying it under the highest ‘critical’ severity tier. However, Oracle assessed the risk at 8.1 on the CVSS 3.1 scale, categorizing the exploit complexity as ‘high’. This is because issues only manifest in environments utilizing specific recent server chips from Intel and AMD, and attackers must precisely match the memory structure for each individual application.

Nevertheless, the security industry remains on high alert due to the relentless onslaught of AI-armed hackers. Last month, Califf used AI to discover a vulnerability (PHP2Shell) in PHP, a web development language powering over two thirds of global websites. Shortly after disclosure, this vulnerability became a target for hackers suspected of having ties to North Korea. According to an attack target list obtained by global security firm Cyfirma, 30 out of 216 targeted servers were located in South Korea.