OpenAI has released GPT-6 Cyber, a security model it rates "Critical" under its own Preparedness Framework, the highest tier for cyber capability and a step above the "High" rating of its predecessor. It is available only to a select group of vetted customers, and OpenAI has set no price or general release date. The announcement on 5 November 2026 makes it the second restricted cyber model in under two weeks, after Google's Gemini 4 Carbon.
What it is and who can use it
GPT-6 Cyber is a variant of OpenAI's GPT-6 Astra built for finding exploitable vulnerabilities and developing fixes, trained to stay within an authorised scope. It is initially available through the API and Codex to selected customers in Daybreak Red, OpenAI's tier for deeper exploit development and authorised security testing. OpenAI says it will expand access to all Daybreak Red customers in the coming weeks. It is also opening the older GPT-5.6 Cyber to all Red customers and widening the Daybreak Blue alpha, which gives verified defenders a version of GPT-6 Astra with fewer unnecessary refusals. A new group, Daybreak Red for Critical Infrastructure, will let approved defenders test critical systems with safeguards against model misalignment.
OpenAI's results
- It scored 48.5% on ExploitGym, an OpenAI evaluation, against 32.1% for GPT-5.6 Cyber.
- It completed 95% of tasks in RemoteExploitBench, which covers known vulnerabilities in applications such as GitLab, PostgreSQL and Jenkins. GPT-6 Astra in its less permissive Blue setting refused every task.
- OpenAI researchers used it to find 81 previously unknown high-severity flaws in Firefox, which Mozilla has begun fixing.
- OpenAI says it has disclosed 304 critical Linux findings to maintainers with proposed patches, building on the 948 Linux root exploits we reported earlier this week.
OpenAI also reports that the model stayed within its authorised scope in every tested case with deployment safety prompts, and in 99.93% of honeypot tests without them. These are OpenAI's own evaluations and have not been independently verified.
Why "Critical" matters
Under OpenAI's framework, Critical is the top risk level for cyber. It shares its ratings with GPT-6 Astra, which is also rated High for biological and chemical risk. The rating is why access is limited: the same capabilities that help defenders find flaws could help attackers if misused. OpenAI says it has refined safeguards against help that could compromise industrial controls or cause serious physical damage.
The concerns
Security experts quoted by The Next Web and Technology.org pointed to two worries. One is that even a small rate of out-of-scope behaviour by a highly capable model could mean actions against systems nobody authorised. The other is a capacity gap: Chris Hughes of Noma Security noted that OpenAI's own Patch the Planet dashboard shows fewer than 25% of the flaws found so far have been fixed, so finding bugs faster does not by itself make software safer.
Why it matters
The pattern is now clear. The most capable security models from the leading AI labs are being released in stages to vetted defenders first, with the public left to wait. For organisations, including governments and businesses across Africa that rely on open-source software, the practical consequence is that vulnerability discovery is speeding up, and patching will need to keep pace. Access to these tools is for approved security teams that meet OpenAI's identity and access requirements.
Sources: OpenAI, "Introducing GPT-6 Cyber" (5 November 2026); The Hacker News; The Next Web; Technology.org; Dataconomy; BigGo Finance.
