OpenAI's GPT-6 Cyber Finds 81 Firefox Vulnerabilities and Develops 948 Linux Exploits

OpenAI has introduced GPT-6 Cyber, a specialised artificial intelligence model designed to help cybersecurity professionals discover software vulnerabilities, test how they could be exploited and develop fixes before attackers can take advantage of them.

The announcement, published on November 5, 2026, comes with figures that highlight how quickly AI-assisted security research is advancing. OpenAI says its researchers used the model to identify 81 previously unknown high-severity vulnerabilities in Mozilla Firefox and develop working exploits for 948 Linux kernel vulnerabilities.

The company also says the model is better than its predecessor at respecting the boundaries of authorised security testing. However, OpenAI has classified GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework, reflecting the potential risks associated with its capabilities.

The launch raises an important question for the technology industry: as AI becomes better at finding software weaknesses, can it help defenders fix those weaknesses faster than attackers can exploit them?

What Is GPT-6 Cyber?

GPT-6 Cyber is a specialised AI model developed for advanced cybersecurity research. Unlike a general-purpose chatbot, it is designed to work on complex security tasks, including analysing software, identifying vulnerabilities, validating whether weaknesses can be exploited and helping researchers develop patches.

These tasks traditionally require considerable technical knowledge and time. Security researchers may need to examine large codebases, trace how different components interact, reproduce suspicious behaviour and determine whether a vulnerability could lead to a real attack.

AI systems with stronger coding and reasoning capabilities can assist with several stages of that process.

OpenAI says GPT-6 Cyber was trained to find and validate exploitable vulnerabilities while remaining within the boundaries set by security researchers. This distinction is important because the same technical capabilities that help defenders test their systems could also be misused to attack systems without permission.

The Firefox Discovery: 81 Previously Unknown High-Severity Vulnerabilities

One of the most significant results highlighted by OpenAI involves Mozilla Firefox.

According to the company, its researchers used GPT-6 Cyber to uncover 81 previously unknown vulnerabilities that Mozilla classified as high severity. The model also helped identify 30 existing issues that were more serious than originally understood.

OpenAI says it shared evidence, reproducible tests and proposed patches with Mozilla. Mozilla has since begun releasing fixes for many of the newly identified vulnerabilities.

The result is notable because Firefox is a widely used browser that has undergone extensive security testing over many years. Finding additional high-severity weaknesses in such a mature software project demonstrates how AI-assisted research could help security teams examine complex applications more thoroughly.

It does not mean that Firefox is broadly unsafe or that every identified vulnerability remains exploitable. The important point is that researchers found previously unknown issues and began coordinating with Mozilla to address them.

GPT-6 Cyber Also Developed 948 Linux Kernel Exploits

The Linux findings are another major part of the announcement.

OpenAI says GPT-6 Cyber reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 of them. The company reports that 588 of those vulnerabilities were previously unknown.

OpenAI also says it disclosed 304 critical findings to Linux maintainers, together with candidate patches, following guidance from the Linux community. Work is continuing to address other reported issues.

The Linux kernel is the core component that allows the operating system to communicate with hardware and manage essential system resources. It is used across servers, cloud infrastructure, embedded devices and many other computing environments.

That makes kernel security particularly important. A serious vulnerability can potentially affect systems that support websites, business applications, cloud platforms and critical digital services.

The figures should be understood in context. Developing an exploit in a research environment is not the same as compromising a live system. The findings demonstrate the model's capabilities in controlled security research, not proof that 948 production systems have been attacked.

A 51% Improvement on OpenAI's ExploitGym Benchmark

OpenAI also published benchmark results comparing GPT-6 Cyber with its predecessor, GPT-5.6 Cyber.

On ExploitGym, an evaluation that tests whether a model can exploit a specified vulnerability, GPT-6 Cyber achieved a score of 48.5%. GPT-5.6 Cyber scored 32.1%.

OpenAI describes this as a 51% relative improvement.

The company also reported better performance on its tests for staying within authorised boundaries. Without its deployment safety prompt, GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of tests. With the prompt, it stayed within scope in all tests in that evaluation.

These results are encouraging, but they need to be interpreted carefully. They are evaluations published by OpenAI, and benchmark performance does not guarantee identical behaviour in every real-world environment.

The figures nevertheless show that the company is working to improve two important aspects of cybersecurity AI: technical capability and the ability to follow clearly defined testing permissions.

Why OpenAI Gives GPT-6 Cyber a Critical Security Rating

OpenAI has assigned GPT-6 Cyber a Critical cybersecurity classification under its Preparedness Framework. Its previous GPT-5.6 Cyber model had a High classification.

The classification reflects the model's ability to perform advanced cybersecurity tasks, including finding vulnerabilities and developing working exploits. These capabilities can help legitimate security teams, but they also create risks if used by people attempting unauthorised attacks.

This is why access to the model is restricted.

OpenAI says it has introduced additional safeguards, including monitoring designed to detect activity outside authorised scope, protections against misuse of credentials and controls for high-risk cybersecurity assistance. It also recommends sandboxing, human oversight and action review when using models in sensitive security workflows.

The company is also paying particular attention to critical infrastructure, including power plants, water treatment facilities and industrial control systems. These environments require stronger protections because a successful cyberattack could disrupt essential services or cause physical harm.

Who Can Access GPT-6 Cyber?

GPT-6 Cyber is not being released as a general consumer chatbot.

At launch, OpenAI made it available through its API and Codex to a select group of approved Daybreak Red customers. Daybreak Red is the company's access tier for advanced authorised security work, including penetration testing and exploit development.

OpenAI says it plans to expand access to all approved Daybreak Red customers over the coming weeks. It is also making GPT-5.6 Cyber available to all Daybreak Red customers.

This controlled rollout allows the company to provide advanced capabilities to vetted security professionals while monitoring how the technology is used.

What This Means for Businesses and Cybersecurity Teams

For businesses, the announcement points to a future in which vulnerability research and software security reviews could become faster and more continuous.

Security teams could use AI assistance to examine large codebases, investigate suspicious behaviour, validate potential weaknesses and prepare fixes. This could be particularly useful for organisations managing complex software environments or maintaining applications with limited security resources.

Software developers may also see AI security tools become a more regular part of the development process. Instead of treating security reviews as a final step before release, teams could increasingly use AI-assisted checks throughout development.

However, AI should not replace professional security testing, secure development practices or human review. Businesses still need patch management, access controls, backups, monitoring, incident response plans and properly authorised penetration testing.

The model's findings also underline why organisations should keep their software updated. When vulnerabilities become easier to discover and validate, delays in applying security fixes can become more consequential.

The Bigger Picture: AI Is Changing Both Sides of Cybersecurity

The development of GPT-6 Cyber reflects a wider shift in the cybersecurity industry.

AI can help defenders investigate weaknesses, understand attack paths and develop fixes. At the same time, increasingly capable models could lower the technical barriers for people attempting to carry out cyberattacks.

That creates a race between vulnerability discovery and remediation. Finding a security flaw is only one part of the process. Organisations must also verify the issue, prioritise its severity, develop a reliable fix and deploy that fix before attackers can take advantage of it.

OpenAI's announcement suggests that advanced AI systems could become valuable tools in this process. But their impact will depend on responsible access, effective safeguards, independent validation and how quickly organisations act on the vulnerabilities they uncover.

For now, GPT-6 Cyber offers a significant indication of where AI-assisted security research is heading. Its ability to identify previously unknown Firefox vulnerabilities and develop Linux kernel exploits demonstrates the growing technical potential of specialised cybersecurity models.

The challenge ahead is ensuring that these capabilities consistently strengthen digital security rather than making existing threats harder to manage.

Frequently Asked Questions

What is OpenAI GPT-6 Cyber?

GPT-6 Cyber is a specialised AI model designed to help authorised cybersecurity professionals discover software vulnerabilities, validate exploits and develop fixes.

How many Firefox vulnerabilities did GPT-6 Cyber find?

OpenAI says the model helped identify 81 previously unknown vulnerabilities in Firefox that Mozilla classified as high severity. It also identified 30 existing issues that were more serious than previously understood.

How many Linux exploits did GPT-6 Cyber develop?

OpenAI reports that the model reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 of them.

Is GPT-6 Cyber available to everyone?

No. At launch, access was limited to selected, approved Daybreak Red customers through OpenAI's API and Codex. The company says it plans to expand access to approved customers.

Why is GPT-6 Cyber classified as Critical?

OpenAI assigned the model its Critical cybersecurity classification because of its advanced ability to discover vulnerabilities and develop working exploits. The classification reflects the potential risks of these capabilities as well as their defensive value.

Does GPT-6 Cyber mean traditional cybersecurity tools are no longer necessary?

No. AI can support security research, but organisations still need professional security teams, patch management, monitoring, access controls, incident response and independent testing.