OpenAI Launches GPT-6 Cyber With Critical Cybersecurity Rating After Autonomous Linux Kernel Research

OpenAI has introduced GPT-6 Cyber, a specialized AI model built to discover software vulnerabilities, validate exploits and develop fixes. The launch comes with an upgraded internal cybersecurity capability classification of Critical, a step up from the High rating assigned to its predecessor, GPT-5.6 Cyber. OpenAI also flagged growing attempts by malicious actors to misuse cybersecurity-focused AI models.

The model's capabilities were demonstrated through a large-scale Linux kernel research project, in which autonomous AI agents reproduced 1,724 vulnerabilities, developed working exploits and candidate patches for 948 distinct privilege-escalation vulnerabilities, and completed the work at approximately $1.2 million in API costs. Of those 948 vulnerabilities, 588 were previously unknown. Separately, GPT-6 Cyber identified 81 previously unknown high-severity Firefox vulnerabilities. The kernel findings and patches are being reviewed and reported to Linux maintainers rather than deployed directly.

GPT-6 Cyber is initially available to select Daybreak Red customers for advanced, authorized security testing, with broader access planned over the coming weeks. Security teams at Mozilla, Palo Alto Networks and Cisco are already using the model. OpenAI is also expanding its Daybreak Blue alpha for GPT-6 Astra, giving more verified defenders access to AI tools for code reviews, vulnerability investigations and remediation. The broader vision is a continuous security pipeline in which AI agents find vulnerabilities, test their impact and prepare fixes at scale under human oversight.