Deployment of GPT-6 Cyber to Vetted Security Professionals
OpenAI has released GPT-6 Cyber, a new artificial intelligence model specifically engineered for cybersecurity applications, according to reports from Channel News Asia and Fortune. The model is designed to assist security researchers in identifying vulnerabilities, testing exploitation methods, and developing patches before malicious actors can utilize similar techniques.
The release comes as the industry faces increasing concerns regarding the misuse of AI by malicious cyber actors. OpenAI noted that it has observed a rise in attempts to misuse its technology, including efforts by groups seeking to distill its models.
Safety Classification and Access Tiers
In a significant shift in its internal safety protocols, OpenAI has assigned GPT-6 Cyber a “Critical” cybersecurity rating under its safety framework. This represents an escalation from the “High” rating previously assigned to the GPT-5.6 Cyber model.
Access to the new model is currently restricted to a small group of vetted customers through Daybreak Red, OpenAI’s specialized access tier intended for authorized security work, such as exploit development and penetration testing. OpenAI stated that access will be expanded to all approved Daybreak Red customers over the coming weeks. Additionally, the company announced that the previous GPT-5.6 Cyber model will be made available to all Daybreak Red customers starting Thursday.
Performance Metrics and Vulnerability Discovery
OpenAI reported several measurable outcomes from the use of GPT-6 Cyber in research environments. According to the company, its researchers utilized the model to identify 81 previously unknown vulnerabilities in the Firefox browser. Mozilla has since classified these findings as high severity and has begun releasing fixes.
The model’s capabilities in operating system security were also highlighted. OpenAI reported that the model assisted in developing working exploits for 948 Linux kernel vulnerabilities. Of these, 588 were previously unknown. Furthermore, the company disclosed 304 critical Linux findings to maintainers, accompanied by proposed patches.