OpenAI unveils AI model 'GPT-6 Cyber' to prevent hacking [Global AI Briefing]

[Financial News] OpenAI has unveiled GPT-6 Cyber, an artificial intelligence (AI) model dedicated to cybersecurity that identifies security vulnerabilities and tests whether they could be exploited in real attacks. The technology goes beyond finding security holes in software: it can generate attack code on its own to assess the risks and help fix them.
After Google recently unveiled Gemini 4 Carbon, a frontier model that identifies hacking vulnerabilities and directly proposes fixes, OpenAI has also introduced an AI tool that helps identify and fix security vulnerabilities, making the security market an emerging target for global AI companies. At the same time, concerns are growing that powerful AI models could be misused as tools by hackers, as they gain the ability to write attack code themselves.
On the 5th (local time), OpenAI announced on its official blog that it had unveiled GPT-6 Cyber, a cybersecurity-focused model trained to analyze software source code, identify security vulnerabilities, test whether they can actually be exploited, and then suggest ways to fix them.

OpenAI said it would initially make GPT-6 Cyber available to users of Daybreak Red, an authorized offensive security partner program, through an application programming interface (API) and Codex.
OpenAI said GPT-6 Cyber had found 81 previously unknown high-risk vulnerabilities in the Firefox web browser. It added that the model had reproduced 1,724 vulnerabilities in the Linux kernel and developed working attack code for 948 of them.
It found new security holes even in software that had undergone security testing over a long period.
OpenAI emphasized that it had "shared the findings with the relevant developers and provided candidate code for fixes."
Mozilla, the developer of Firefox, has already released fixes for some of the vulnerabilities.
OpenAI emphasized that GPT-6 Cyber was the first of its products to be assigned a 'Critical' rating—the highest risk level—in the cybersecurity risk assessment under its Preparedness Framework.
OpenAI explained, "That is because it is a defensive tool that finds and fixes weaknesses before hackers do, but if misused, it carries the risk of causing enormous damage, including paralyzing computer networks."
Ultimately, OpenAI itself has effectively acknowledged that its cybersecurity-focused AI model could become a powerful hacking tool if misused.
Industry experts say that improvements in AI companies' cybersecurity-focused models could prove to be a 'double-edged sword,' affecting both cyberattack threats and defensive capabilities in the global security market.
[email protected] Lee Gu-sun Reporter