OpenAI launches GPT-6 Cyber, an AI that finds hidden flaws in software

OpenAI has officially introduced GPT-6 Cyber, a new artificial intelligence model designed to identify vulnerabilities in computer systems. The system has received a "Critical" risk classification according to the company's internal guidelines, highlighting its advanced capabilities and the potential dangers associated with possible misuse.

During initial analyses, the model identified 81 previously unknown vulnerabilities in Mozilla Firefox. In a mid-October test on the Linux kernel, it demonstrated 948 flaws with an equal number of working attacks, through which a regular user already inside the computer can become an administrator. The entire Linux testing campaign cost approximately $1.21 million in computing power, with an average expense of $1,279 for each exploit developed.

OpenAI stated that GPT-6 Cyber successfully executed ten novel attack techniques, never before observed by specialists. To limit potential misuse, the company has introduced several protective measures, including network access restrictions and automated monitoring systems.

GPT-6 Cyber represents a technological leap in the field of cybersecurity, a sector that traditionally requires years of experience and specialized knowledge to identify vulnerabilities in software. Before the advent of these advanced AI tools, finding security flaws was a manual and time-consuming process, often entrusted to human experts who analyzed thousands of lines of code.

The model stems from the evolution of generative artificial intelligence systems applied to cybersecurity. OpenAI developed this specialized tool following the security incident that occurred at Hugging Face, which highlighted the need for more effective tools to protect systems. To test its capabilities, the company created a dedicated environment with honeypots around assigned targets, verifying that the AI did not attempt to access unauthorized resources.

The implications are significant: on one hand, tools like GPT-6 Cyber can strengthen the protection of computer systems by reducing costs and time; on the other hand, they raise concerns about security if they were to fall into the wrong hands. For this reason, OpenAI has also established a specialized group focused on defending strategic infrastructure such as power plants and water services, while the model remains accessible only to selected partners under strict human oversight.