OpenAI's GPT-6 Cyber Finds Security Flaws Before Hackers

OpenAI has introduced GPT-6 Cyber, an experimental AI model designed to identify cybersecurity vulnerabilities and test how they could potentially be exploited in cyberattacks. The company has classified the model's cybersecurity risk level as Critical, higher than the High rating assigned to its previous GPT-5.6 Cyber model. This reflects both its advanced capabilities and the potential risks of misuse.

In this video, we explore what GPT-6 Cyber can do:

• Advanced Vulnerability Detection: The model is designed to identify security weaknesses in software and computer systems.
• Exploit Testing: Authorized cybersecurity researchers can examine how identified vulnerabilities might be exploited, helping them address risks before attackers do.
• Firefox Security Findings: OpenAI says its researchers used the model to identify several vulnerabilities in Mozilla Firefox, which Mozilla classified as high-risk and is working to address.
• Linux Kernel Vulnerabilities: OpenAI reported 304 critical security vulnerabilities to Linux kernel maintainers and proposed fixes.
• Restricted Access: The model is initially available to selected customers through OpenAI's Daybreak Red program, intended for advanced and authorized cybersecurity work.

GPT-6 Cyber highlights the growing role of AI in defensive cybersecurity. However, its ability to identify and test vulnerabilities also raises concerns about how such technology could be misused.