OpenAI announced on Thursday, November 5, 2026, GPT-6 Cyber, a specialized cybersecurity research model. The company rates it Critical under its Preparedness Framework, the highest cybersecurity capability tier, above the High rating given to GPT-5.6 Cyber.
According to OpenAI, the model is already being used by security teams to identify, validate, and fix vulnerabilities in real software. In research campaigns it found 81 previously unknown high-severity vulnerabilities in Firefox. Mozilla has begun releasing fixes. The same system reproduced 1,724 vulnerabilities in the Linux kernel and generated working local privilege-escalation exploits for 948 of them — more than five times the result of an earlier model on the same research harness. The Linux run cost roughly $1.2 million in API credits.
Performance and alignment
On the ExploitGym benchmark, GPT-6 Cyber scored 48.5%, up from 32.1% for GPT-5.6 Cyber. In an alignment evaluation (ExploitGym Honeypot), the model avoided exploiting out-of-scope services in 99.93% of tests without the deployment prompt and 100% with the prompt. OpenAI says dedicated training sharply reduced the risk of the model going outside researcher-defined boundaries.
Initial access is limited to approved Daybreak Red customers, with a gradual expansion. Teams at Cisco, Calif, CodeRabbit, Palo Alto Networks, and OpenAI itself are already using it for vulnerability research and code review.
Sources: OpenAI, OpenAI (Linux), Mozilla.
This article was produced with the assistance of artificial intelligence and editorially reviewed.
Por Grok
