OpenAI introduces GPT-6 Cyber, a security research model it says pairs stronger capabilities with better model alignment. Select approved Daybreak Red customers can use it now, and OpenAI says access will expand over time.

What it has been used for

OpenAI says GPT-6 Cyber powered the Linux kernel research it announced the day before, and that security teams at Mozilla, Palo Alto Networks, Calif, Cisco and OpenAI itself are already putting it to work. Fortune reports OpenAI's figures that researchers used the model to find 81 previously unknown high-severity Firefox vulnerabilities and to develop working exploits for 948 Linux kernel vulnerabilities, 588 of which were previously unknown.

In an initial RemoteExploitBench evaluation, Unite.ai reports that GPT-6 Cyber completed 95% of tasks using known flaws in GitLab, PostgreSQL and Jenkins in a controlled environment, while GPT-6 Astra refused every task. The same report says those results do not show reliable exploitation in every setting. The system card gives an intended-exploit success rate of 48.45% for GPT-6 Cyber against 42.43% for GPT-6 Astra.

Who can get it and how

Access runs through Daybreak Red. OpenAI says the model is available to select Daybreak Red customers and that it is expanding access over time. The system card says Daybreak Red requires identity verification, use restrictions, single sign-on, multi-factor authentication and API IP allowlisting.

OpenAI describes two tiers. Daybreak Blue helps verified defenders validate vulnerabilities and fixes with fewer refusals. Daybreak Red enables offensive security work such as deeper exploit development and authorized testing. Crypto Briefing reports that Daybreak launched in May 2026 and split into the two tiers on August 10-11, 2026.

OpenAI is also widening the Daybreak Blue alpha for GPT-6 Astra to more organizations and individual researchers. Its stated goal is to extend Daybreak access to organizations of every size, independent researchers and open-source maintainers. For background on GPT-6 Astra, see our earlier report on the model.

Safety controls and a data-use change

The system card says extra checks were added to GPT-6 Cyber and cyber-critical Daybreak Blue models to detect and block high-risk help involving industrial controls, essential services and election systems. Those checks were evaluated with the industrial control security firm Dragos. Windows Report says OpenAI classifies GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber.

Pasquale Pillitteri reports that in an internal OpenAI test called ExploitGym Honeypot, GPT-6 Cyber avoided attacking out-of-scope services 99.93% of the time without safety prompts and 100% with them, against 94.49% for GPT-5.6 Cyber.

CyberScoop reports that OpenAI told Daybreak Blue members that session data from personal accounts would be used to improve models starting at the end of the month, even for users who had opted out. OpenAI later said customers can keep opting out of sharing Daybreak usage for training.

Our read: the access controls matter as much as the benchmark scores, since the benefit to defenders depends on who gets through verification.

What's next

OpenAI has scheduled a Cyber at the Frontier livestream for November 12 at 10am PT, promising more Cyber announcements.

Sources