The short version
● What — GPT-6 Cyber, a security-focused model
● When — 5 November 2026
● Rating — Critical for cybersecurity, up from High
● Access — vetted Daybreak Red customers only
OpenAI launched GPT-6 Cyber on 5 November. It scored 48.5% on ExploitGym, which measures whether a model can exploit a flaw it has been handed, against 32.1% for GPT-5.6 Cyber. On Firefox it surfaced 81 previously unknown high-severity flaws, and Mozilla has started shipping fixes.
The Linux numbers are larger. With the same setup, GPT-6 Cyber reproduced 1,724 vulnerabilities and built working exploits for 948 of them; GPT-5.6 Sol managed 166. OpenAI has disclosed 304 critical Linux findings to maintainers along with proposed patches. On a Dragos mock power plant it found two zero-days in engineering software handling protective relay data, and on a remote attack benchmark it completed 95% of tasks where GPT-6 Astra refused every one.
OpenAI rated the model Critical for cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber, and access is restricted to vetted customers in Daybreak Red, its tier for exploit development and authorised testing, reachable through the API and Codex. On the scope question, without its deployment safety prompt the model stayed out of unauthorised services in 99.93% of ExploitGym Honeypot tests against 94.49% for its predecessor; with the prompt, it stayed in bounds every time. Our generation-on-generation comparison puts every one of those numbers side by side.
More from 6 November
Full dated index: every AI release in November 2026.
Sources