OpenAI announced on November 5, 2026, GPT-6 Cyber, a specialized variant of GPT-6 Astra designed for authorized cybersecurity research. The model is rated Critical under the company’s Preparedness Framework — the highest cybersecurity capability tier — and is initially available only to approved customers in the Daybreak Red program.
What the model does
According to OpenAI, GPT-6 Cyber was trained to find and validate exploitable vulnerabilities, develop exploits in controlled environments, and stay within the researcher-authorized scope. In internal benchmarks it scored 48.5% on ExploitGym (versus 32.1% for GPT-5.6 Cyber). In alignment tests it avoided exploiting out-of-scope services in 99.93% of runs without a safety prompt and 100% with the prompt.
The company reported that researchers used the model to identify 81 previously unknown high-severity vulnerabilities in Firefox (Mozilla has begun releasing fixes) and to develop working exploits for 948 Linux kernel vulnerabilities, including hundreds that were previously unknown.
Why it matters
The advance highlights the growing ability of AI models on offensive security tasks while OpenAI stresses improved alignment to reduce out-of-scope risk. Access remains restricted with identity verification, multi-factor authentication, and organizational controls. It intensifies the debate on accelerating cyber defense without excessively expanding misuse potential.
By Grok
