A cybersecurity start-up has identified a critical vulnerability in Java, a widely used programming language, through the use of advanced artificial intelligence (AI) tools. The discovery, announced Tuesday, underscores growing challenges and opportunities in cybersecurity as AI rapidly transforms the identification of software flaws.
The start-up, Calif, employed OpenAI’s GPT-6 Cyber model to detect the flaw in Java’s base64 decoder—a core component that converts binary data into readable text. Researchers found that on certain systems, processing a specific sequence of characters causes the decoder to malfunction, leading to improper code execution and potential system crashes. By leveraging AI-assisted techniques such as “heap grooming,” the team demonstrated how the flaw could enable unauthorized access to sensitive systems.
Java is integral to many sectors, including banking, healthcare, aviation, technology, and government agencies with national security responsibilities. Bruce Dang, a senior researcher at Calif, described the language as “the beating heart of the modern economy,” emphasizing the widespread impact the vulnerability could have across major companies and institutions.
In the days following the discovery, Calif coordinated disclosures with members of the Five Eyes intelligence alliance—comprising the United States, Canada, Australia, New Zealand, and Britain—and with private sector companies such as Google and Microsoft. Both tech giants confirmed that they deployed mitigations promptly. Eduardo Vela Nava, an information security engineer at Google, stated that the company had quickly rolled out patches and developed detection measures to prevent exploitation.
Oracle, the owner of the Java trademark and licensing authority, issued a software patch on Tuesday to address the security gap but did not provide additional comment. The U.S. National Security Agency (NSA), which experienced a network outage on the same day attributed to routine maintenance, did not confirm any connection to the vulnerability.
The Java flaw, dubbed “Java4Shell,” marks Calif’s second major AI-assisted software vulnerability in recent months. Earlier, the company revealed significant security issues in PHP, a language powering a large portion of websites worldwide. Although the PHP discovery triggered initial concerns about widespread disruptions, the resulting impact was less severe than feared, as hackers found it difficult to exploit the vulnerability.
Experts note that AI’s increasing role in uncovering vulnerabilities allows for faster identification of latent weaknesses but also raises the stakes in cybersecurity. Ollie Whitehouse, chief technology officer of Britain’s National Cyber Security Center, said AI has been responsible for uncovering a growing number of hidden flaws in software that supports essential digital services.
Cybersecurity professionals now face the challenge of containing risks posed by “centaur hackers”—human operators augmented by AI—which can develop sophisticated cyberattacks much faster than before. Comparisons have been drawn to the world of chess, where human-AI hybrids once outperformed both solo players and machines but were eventually surpassed by more advanced AI systems.
Fouad Matin, who leads cyber preparedness at OpenAI, highlighted the positive potential of AI-assisted research to strengthen software security through collaboration. However, some experts caution that the technology may soon allow AI to independently compose highly advanced attacks, escalating the complexity and unpredictability of cyber threats.
The FBI recently issued a warning about autonomous AI hacking tools being deployed against governments and businesses globally, underscoring the urgency of adapting cybersecurity strategies. As one researcher summarized, the era of AI-enhanced cyberattacks is already here, and the window to prepare for fully autonomous AI threats may be limited to months or a few years ahead.
