OpenAI Introduces a New Generation of AI-Powered Cybersecurity
OpenAI has launched GPT-6 Cyber, a specialised artificial intelligence model developed to identify cybersecurity vulnerabilities, test potential attacks and help security researchers fix software weaknesses.
Announced on November 5, 2026, the model represents a significant development in AI-powered cybersecurity. OpenAI has assigned it a Critical cybersecurity capability rating under its Preparedness Framework, moving up from the High rating given to its previous GPT-5.6 Cyber model.
The model is designed primarily for authorised security professionals who need advanced tools to investigate software vulnerabilities, conduct penetration testing and improve the security of digital infrastructure.
Unlike general-purpose AI assistants, GPT-6 Cyber has been specifically trained to handle complex cybersecurity tasks, including vulnerability discovery and exploit development.
Why Has OpenAI Given GPT-6 Cyber a Critical Rating?
The Critical classification reflects the model’s advanced ability to identify and exploit software vulnerabilities.
This rating does not mean the model itself is unsafe or that it has been released for unrestricted hacking. Instead, it indicates that its cybersecurity capabilities have reached a level requiring stronger safeguards and controlled access.
According to OpenAI, GPT-6 Cyber delivers substantial improvements over its predecessor.
On the ExploitGym benchmark, which evaluates a model’s ability to exploit a specified vulnerability, GPT-6 Cyber achieved a score of 48.5%, compared with 32.1% for GPT-5.6 Cyber.
It also demonstrated improved ability to identify vulnerabilities in web applications and conduct more complex security research.
The model is intended to help defenders discover weaknesses before malicious actors can take advantage of them.
GPT-6 Cyber Helped Discover 81 Firefox Vulnerabilities
One of the most significant results highlighted by OpenAI involves Mozilla Firefox.
OpenAI researchers used GPT-6 Cyber to identify 81 previously unknown vulnerabilities in Firefox that Mozilla classified as high severity.
The model also helped researchers develop evidence and reproducible tests for the discovered issues, along with proposed patches to help Mozilla engineers address them.
Mozilla has started releasing fixes for many of these vulnerabilities.
The achievement is particularly important because Firefox is a widely used browser that has undergone extensive security testing over many years.
Finding previously unknown high-severity weaknesses in such established software demonstrates the potential of AI-assisted vulnerability research.
Nearly 1,000 Linux Kernel Vulnerabilities Exploited During Research
OpenAI also reported substantial results from cybersecurity research involving the Linux kernel, which powers a significant portion of the world’s servers, cloud infrastructure and computing systems.
According to the company, GPT-6 Cyber helped researchers reproduce 1,724 vulnerabilities and develop working exploits for 948 of them.
The findings included 588 vulnerabilities that were previously unknown.
OpenAI said it disclosed 304 critical findings to Linux maintainers and provided candidate patches to support the process of fixing the weaknesses.
The research demonstrates how advanced AI models could help security teams investigate large and complex codebases more efficiently.
However, it also highlights why powerful vulnerability discovery tools must be carefully controlled: the same capabilities that help defenders can potentially be misused by attackers.
Stronger Protection Against Unauthorised Cyber Activity
OpenAI says improving cybersecurity capabilities was only one part of developing GPT-6 Cyber. The company also focused on making the model more reliable at following authorisation boundaries.
A major concern with autonomous cybersecurity agents is that they may move beyond the systems they have been authorised to test.
For example, an agent assigned to examine one application should not attempt to access unrelated servers, credentials or connected services.
OpenAI reports that GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of tests without deployment safety prompts. With those prompts enabled, it stayed within the authorised scope in 100% of the evaluated tests.
The company has also introduced additional safeguards, including:
- Cybersecurity-specific monitoring to identify potentially unauthorised activity.
- Restrictions on access to advanced cybersecurity capabilities.
- IP allowlisting requirements for Daybreak Red API access by default.
- Additional monitoring for compromised credentials.
- Sandboxing and human oversight for high-risk workflows.
- Stronger protections against misuse involving critical infrastructure.
These measures are intended to make the model more useful for legitimate security research while reducing the possibility of unauthorised attacks.
GPT-6 Cyber Will Initially Be Available to Selected Security Professionals
OpenAI is not releasing GPT-6 Cyber as a generally available chatbot.
The model is initially being offered to a limited group of approved customers through Daybreak Red, the company’s access programme for advanced and authorised cybersecurity work.
Access is designed for security professionals and organisations conducting activities such as penetration testing, vulnerability research and exploit development.
OpenAI plans to expand access to all approved Daybreak Red customers over the coming weeks.
The company is also making its earlier GPT-5.6 Cyber model available to all Daybreak Red customers.
This restricted rollout reflects the sensitivity of the model’s capabilities and the need to verify that users have legitimate security requirements.
Protecting Critical Infrastructure Is a Major Priority
One of the most important areas of concern for OpenAI is the protection of critical infrastructure.
Power plants, water treatment facilities, industrial control systems and other essential services rely on interconnected digital technologies. A successful cyberattack against these systems could create serious operational and public safety consequences.
OpenAI says it has introduced additional safeguards to prevent unauthorised assistance involving industrial controls and other safety-critical systems.
At the same time, the company wants authorised infrastructure defenders to benefit from advanced AI tools.
Its Daybreak programme includes initiatives intended to expand access for approved defenders, including government organisations and essential service providers.
OpenAI has also worked with cybersecurity specialists to evaluate how its models perform in controlled industrial environments.
The goal is to support legitimate security testing without enabling activities that could disrupt essential services or cause physical damage.
What GPT-6 Cyber Means for the Future of Cybersecurity
The launch of GPT-6 Cyber highlights a major shift in the cybersecurity industry.
Traditional security research often requires specialists to manually inspect code, investigate vulnerabilities, reproduce attacks and develop patches. Advanced AI systems can increasingly assist with several of these stages.
For businesses, this could mean faster vulnerability detection, more efficient code reviews and improved protection against emerging threats.
However, the technology also creates new challenges. Attackers may use increasingly capable AI systems to discover vulnerabilities and develop exploits more quickly.
This makes timely patching, strong access controls, continuous monitoring and responsible AI deployment increasingly important.
AI Cybersecurity Is Entering a New Competitive Phase
OpenAI’s GPT-6 Cyber launch reflects the growing competition to develop AI systems that can defend increasingly complex digital environments.
The company’s research results suggest that AI can identify weaknesses even in heavily scrutinised software projects and help developers produce fixes.
But the Critical cybersecurity classification also underlines the responsibility that comes with these capabilities.
As AI models become more capable of understanding software and developing working exploits, the cybersecurity industry will need to ensure that defensive access grows alongside stronger protections against misuse.
The future of cybersecurity may increasingly depend on how effectively organisations use advanced AI to discover and repair vulnerabilities before attackers can exploit them.
