OpenAI has introduced a new artificial intelligence model designed to assist cybersecurity professionals, saying the system—called GPT-6 Cyber—outperforms its general-purpose GPT-6 Astra model on most cyber tasks. Because the model could be used to conduct cyberattacks if it fell into the wrong hands, OpenAI is currently making it available only to a small group of approved security customers.

Fortune first reported in September that OpenAI was planning to release a specialized cyber model built on GPT-6.

The launch comes as AI models grow increasingly adept at discovering software flaws and turning them into working attacks, putting businesses under mounting pressure to reinforce their cyber defenses. In a September forecast, Gartner projected that worldwide AI cybersecurity spending would reach $51.3 billion this year, nearly double last year's $25.9 billion. OpenAI is leaning further into this market, viewing it as an important path to growing its enterprise sales. GPT-6 Cyber is the company's fourth cybersecurity-focused model this year, following versions released in April, June, and August. OpenAI has also hired a chief revenue officer, Dali Rajic, who comes from a cybersecurity software background.

Rivals are competing for the same customers. Anthropic's Glasswing program gives vetted users access to capabilities that its public models might otherwise block. Google, which previously emphasized a lightweight, low-cost model for repeatedly scanning large codebases, has begun offering vetted defenders access to its frontier Gemini 4 Argon model without cyber guardrails through its Fairwind Program.

GPT-6 Cyber is focused on advanced vulnerability research and exploit development. To test a server's defenses, researchers may attempt to break into it to see what an attacker could do with a given flaw. The same techniques can be used in malicious attacks, so general-purpose AI models impose restrictions that can inadvertently block legitimate security work. OpenAI has said it trains its Cyber models both to sharpen specialized cybersecurity skills and to reduce refusals for certain tasks that can serve attackers and defenders alike.

OpenAI evaluated GPT-6 Cyber on two heavily hardened targets: Firefox, Mozilla's web browser, which must handle potentially malicious webpages without exposing users' data, and the Linux kernel, the core software underlying many servers, which controls access to a machine's resources. Both are open-source projects whose code has been inspected, tested, and hardened by developers and security researchers for years—yet AI models are still able to find novel vulnerabilities.

According to OpenAI, researchers used GPT-6 Cyber to identify 81 previously unknown, high-severity Firefox vulnerabilities and to develop working exploits for 948 Linux kernel vulnerabilities, including 588 previously unknown flaws. The Linux exploits could allow an attacker with ordinary local access to gain root privileges, the highest level of control over a system. Researchers developed working exploits for more than five times as many Linux vulnerabilities as in a July study using GPT-5.6 Sol, which was regarded as a state-of-the-art model for cyber tasks when it was released that month. Results like these are what the emerging market for vetted-access cyber AI is built around: capabilities valuable enough to accelerate defensive research, and sensitive enough that vendors are restricting who can use them.

OpenAI currently offers two tiers of access to trusted cybersecurity professionals through its Daybreak program. Daybreak Red supports advanced work such as simulating attacks and building tools that enable red-teaming, while Daybreak Blue focuses on defensive tasks such as reviewing code, analyzing malware, and investigating incidents. GPT-6 Cyber will initially be available to a select group of approved Daybreak Red customers only, with access expanding to all Red customers over the coming weeks, the company said. Separately, OpenAI said it is expanding early testing of Astra for defensive security work to more organizations and individual security researchers through Daybreak Blue. With access set to widen over the coming weeks, the staged rollout will offer the first broad look at how specialized cyber models are put to work in day-to-day security operations.

This story was originally featured on Fortune.com.