OpenAI launched GPT-6 Cyber, a Daybreak Red cybersecurity model it rates Critical under its Preparedness Framework, first for a limited set of approved customers.

OpenAI launched GPT-6 Cyber, a Daybreak Red cybersecurity model it rates Critical under its Preparedness Framework, first for a limited set of approved customers.

"We’re introducing GPT‑6 Cyber, a major leap in security research: more capable at finding and fixing vulnerabilities, and far more reliable staying within the boundaries researchers set," OpenAI wrote on Thursday (OpenAI). On ExploitGym it scores 48.5%, up from 32.1% for GPT‑5.6 Cyber. In a separate alignment test, it "avoided exploiting out-of-scope services in 99.93% of tests without its deployment prompt, and 100% with this prompt." OpenAI classifies the model as "Critical in cybersecurity under our Preparedness Framework, up from High for GPT‑5.6 Cyber." It says researchers used the model to identify "81 previously unknown high-severity vulnerabilities in Firefox" (Mozilla has been landing fixes) and to develop "948 privilege escalation exploits in Linux" — the Linux pipeline we covered yesterday. Access is gated: "GPT‑6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers," with expansion to all Daybreak Red customers planned, plus GPT‑5.6 Cyber for all Daybreak Red. OpenAI also points to a system card and to new Daybreak Red for Critical Infrastructure permissions for approved defenders. This is a product for vetted security teams, not a general ChatGPT feature.

What to weigh: unless you’re already in Daybreak Red (or apply and get approved), you won’t call this model yourself. The practical signal for everyone else is that OpenAI is shipping a Critical-rated cyber specialist with much tighter "stay in scope" numbers than its predecessor, while keeping it behind an approval wall.