OpenAI GPT-6 Cyber: Finding 81 Firefox & 948 Linux Exploits

OpenAI has introduced a specialised artificial intelligence model that could change how software vulnerabilities are discovered, tested and fixed. Called OpenAI GPT-6 Cyber, the model has helped researchers identify 81 previously unknown high-severity vulnerabilities in Mozilla Firefox and develop working exploits for 948 Linux kernel vulnerabilities.

But the numbers are only part of the story. OpenAI has assigned GPT-6 Cyber its highest cybersecurity risk classification, Critical, under the company’s Preparedness Framework. The classification reflects the model’s growing ability to discover and exploit software weaknesses, raising important questions about how AI-powered security research could change the balance between cyber defenders and attackers.

OpenAI says GPT-6 Cyber is designed for authorised security research, including vulnerability discovery, exploit validation and patch development. Its initial availability is restricted to approved security customers rather than ordinary ChatGPT users.

Here’s what the model can do, what the Firefox and Linux findings actually mean, and why its Critical rating matters.

What Is OpenAI GPT-6 Cyber?

GPT-6 Cyber is a specialised cybersecurity model developed by OpenAI to help security researchers identify software weaknesses, determine whether vulnerabilities can be exploited and develop fixes.

Unlike a general-purpose AI assistant, it is designed for complex security workflows. These can include analysing large codebases, connecting multiple weaknesses, reproducing vulnerabilities and developing proof-of-concept exploits in controlled environments.

The model is also intended to operate within researcher-defined boundaries. This means testing should remain limited to systems and services included in an authorised engagement.

This distinction is important because the same technical capabilities that help defenders identify vulnerabilities could also be misused by attackers.

GPT-6 Cyber Found 81 Previously Unknown Firefox Vulnerabilities

One of the most significant results came from Mozilla Firefox, a browser used by millions of people worldwide.

OpenAI says its researchers used GPT-6 Cyber to identify 81 previously unknown vulnerabilities in Firefox. Mozilla classified these findings as high severity. OpenAI supplied evidence, reproducible tests and proposed patches to help Mozilla investigate and address them.

The research also identified 30 previously known issues that were more severe than originally understood.

This is an important distinction: AI-assisted security research is not limited to discovering entirely new bugs. It can also help security teams reassess existing weaknesses and understand their potential impact more accurately.

Firefox is a particularly interesting example because its codebase has already received extensive security testing from developers and independent researchers.

The findings demonstrate how AI-assisted analysis can potentially identify weaknesses that remain difficult to find through conventional testing alone.

However, discovering 81 vulnerabilities does not mean that all Firefox installations were successfully compromised. Vulnerability discovery, exploitability and real-world exploitation are different stages of a security incident.

948 Working Linux Exploits: Why This Number Matters

The Linux kernel findings are another major part of OpenAI’s announcement.

Linux powers a huge range of technology, including cloud infrastructure, enterprise servers, Android devices, networking equipment and embedded systems. Weaknesses in the kernel can therefore have consequences beyond a single application.

According to OpenAI, GPT-6 Cyber helped researchers:

• Reproduce 1,724 Linux kernel vulnerabilities.
• Develop 948 working privilege-escalation exploits.
• Disclose 304 critical findings to Linux maintainers, together with candidate patches.

The distinction between the first two figures is important. The model reproduced 1,724 vulnerabilities, but developed working exploits for 948 of them. These numbers should not be interpreted as 948 newly discovered Linux vulnerabilities.

Privilege escalation is a security technique through which an attacker attempts to obtain permissions beyond those initially available. For example, someone who gains limited access to a system may try to exploit a kernel vulnerability to obtain higher-level privileges.

If successful, this can turn a limited compromise into a much more serious security incident.

OpenAI says its research team followed the Linux community’s disclosure guidance and shared critical findings with maintainers, including candidate patches. The findings are part of the broader process of validation, disclosure and remediation.

Why Has OpenAI Given GPT-6 Cyber a Critical Risk Rating?

The word Critical in this announcement deserves careful explanation.

OpenAI has not classified the 81 Firefox vulnerabilities as Critical. Instead, it has assigned GPT-6 Cyber itself a Critical cybersecurity risk rating under its Preparedness Framework.

The rating reflects the model’s capabilities in areas such as vulnerability discovery, exploit development and remote exploitation.

OpenAI says GPT-6 Cyber achieved 48.5% on ExploitGym, a benchmark that tests whether a model can exploit a specified vulnerability in a remote target.

For comparison, GPT-5.6 Cyber scored 32.1%. That represents a substantial improvement over the previous model’s result.

The company also tested the model on remote exploitation tasks involving known vulnerabilities in applications such as GitLab, PostgreSQL and Jenkins. GPT-6 Cyber completed 95% of these tasks in OpenAI’s initial evaluation, while GPT-6 Astra refused all of them.

These results indicate a significant improvement in the model’s ability to carry out complex cybersecurity tasks. However, controlled benchmark results should not be treated as a measurement of success against arbitrary real-world systems.

Can GPT-6 Cyber Stay Within Authorised Security Testing?

A powerful cybersecurity AI must do more than find vulnerabilities. It must also respect the boundaries of a security engagement.

For example, a researcher may authorise testing of one application while explicitly prohibiting access to connected services, databases or other infrastructure. An AI agent that ignores those restrictions could create security incidents even during legitimate testing.

OpenAI says it specifically trained GPT-6 Cyber to respect researcher-defined scope and tested this behaviour through an evaluation called ExploitGym Honeypot.

The company reported the following results:

The scope figures are encouraging, but they do not establish that the model can never act outside its authorised boundaries. They represent results from a particular evaluation environment, not a guarantee for every deployment.

At the scale of automated security testing, even a small failure rate can matter. Organisations still need access controls, monitoring, clear testing permissions and human oversight.

Who Can Use GPT-6 Cyber?

GPT-6 Cyber is not being released as a standard ChatGPT feature for everyone.

OpenAI says initial access is limited to selected, approved customers through its API and Codex. The model is being made available through Daybreak Red, the part of OpenAI’s cybersecurity programme intended for deeper exploit development and authorised security testing.

The company has indicated that access will expand to additional approved Daybreak Red customers.

OpenAI’s broader Daybreak programme also includes security-focused capabilities for defenders. The aim is to make advanced AI-assisted vulnerability research more accessible to organisations that need to secure their software and infrastructure.

This restricted rollout is significant because a model capable of developing working exploits needs more careful deployment controls than a conventional coding assistant.

What Does GPT-6 Cyber Mean for Cybersecurity?

GPT-6 Cyber highlights a change in the speed and economics of vulnerability research.

Traditionally, discovering and validating complex software vulnerabilities requires specialised researchers who understand programming languages, operating systems, memory management, application architecture and exploitation techniques.

AI models could help automate parts of this work, allowing security teams to examine larger codebases and investigate more potential weaknesses.

1. Software Developers May Find Vulnerabilities Earlier

AI-assisted code review could help developers detect weaknesses during development instead of waiting for external researchers or attackers to discover them.

This could be particularly useful for large open-source projects that have limited resources for security testing.

2. Open-Source Maintainers Face Greater Pressure

The Linux and Firefox findings demonstrate how AI can help examine mature software that has already received years of security attention.

As vulnerability discovery becomes faster, maintainers may face a larger volume of reports that need to be reproduced, prioritised and fixed.

The challenge will not simply be finding bugs. It will also be processing credible reports without overwhelming development teams.

3. Attackers Could Also Benefit From Similar Capabilities

The same techniques used to identify vulnerabilities and develop exploits could potentially be misused.

A tool that reduces the technical effort needed to reproduce vulnerabilities could lower barriers for less experienced attackers. This is one reason why access restrictions and safeguards matter.

4. Security Testing Could Become More Continuous

Instead of relying only on periodic penetration tests, organisations may increasingly use AI-assisted agents to examine code and infrastructure continuously.

Human security professionals would still need to verify findings, decide which risks matter most and approve changes to production systems.

What Should Businesses and Website Owners Do Now?

The GPT-6 Cyber announcement does not mean that every website or Linux server is suddenly vulnerable. It does, however, reinforce the importance of maintaining software and responding quickly to credible security advisories.

Businesses and website administrators should focus on practical security measures:

• Keep software updated: Apply security updates for operating systems, browsers, server software, frameworks and applications.
• Monitor security advisories: Follow official announcements from software vendors and open-source maintainers.
• Review access permissions: Use least-privilege access so compromised accounts or services cannot automatically obtain administrator-level permissions.
• Test systems responsibly: Conduct vulnerability assessments only on systems for which you have explicit authorisation.
• Maintain backups: Keep tested, protected backups so systems can be recovered after a security incident.
• Monitor suspicious activity: Review server logs, authentication events and unexpected privilege changes.
• Verify vulnerability reports: Do not assume every AI-generated security alert is a confirmed vulnerability. Reproduce and validate findings before making changes.

For WordPress website owners, keeping WordPress core, themes, plugins and the underlying PHP environment updated remains essential. A secure website requires attention to the complete hosting stack, not just its visible content management system.

GPT-6 Cyber vs Traditional Cybersecurity Tools

GPT-6 Cyber does not make conventional security tools obsolete. It represents another layer that can assist researchers with more complex investigations.

The most effective approach is likely to combine AI-assisted research with established security testing, independent verification and responsible disclosure.

Frequently Asked Questions

What is GPT-6 Cyber?

GPT-6 Cyber is OpenAI’s specialised cybersecurity model designed to discover and validate software vulnerabilities, develop exploits for authorised testing and help researchers create fixes.

How many Firefox vulnerabilities did GPT-6 Cyber find?

OpenAI says the model identified 81 previously unknown Firefox vulnerabilities classified as high severity by Mozilla. It also identified 30 previously known issues that were more severe than originally reported.

Did GPT-6 Cyber create 948 Linux exploits?

Yes. OpenAI says the model reproduced 1,724 Linux kernel vulnerabilities and developed working privilege-escalation exploits for 948 of them. It also helped disclose 304 critical findings to maintainers with candidate patches.

Why is GPT-6 Cyber rated Critical?

OpenAI assigned the model a Critical cybersecurity risk rating under its Preparedness Framework because of its advanced vulnerability discovery and exploitation capabilities. The rating applies to the model’s capabilities, not to the severity of the Firefox vulnerabilities.

Can anyone access GPT-6 Cyber?

No. Initial access is restricted to selected, approved customers through OpenAI’s API and Codex as part of its Daybreak Red programme.

Does GPT-6 Cyber mean Linux and Firefox are unsafe?

No. Finding vulnerabilities does not automatically mean that every installation is exploitable or compromised. Mozilla and Linux maintainers are involved in the disclosure and remediation process. Users should install relevant security updates when they become available.

The Bigger Picture: AI Is Changing Both Sides of Cybersecurity

GPT-6 Cyber is an important development because it demonstrates how AI can move beyond identifying suspicious code and assist with the more demanding work of reproducing vulnerabilities and developing working exploits.

The discovery of 81 high-severity Firefox vulnerabilities and 948 working Linux privilege-escalation exploits shows the potential scale of AI-assisted security research. OpenAI’s Critical classification also signals that the company considers these capabilities powerful enough to require additional safeguards and controlled access.

The real test will be what happens next. If developers and open-source maintainers can use AI to discover and fix weaknesses faster than attackers can exploit them, the technology could significantly improve software security. If vulnerability discovery accelerates without equally effective patching and responsible deployment, the risks could grow just as quickly.

For now, GPT-6 Cyber offers a clear message: cybersecurity is entering an era in which AI will increasingly help find the weaknesses hidden inside the software that powers the digital world.

Follow The Smart Innovator™ for more such cover stories. Subscribe to our Newsletters for tech world updates. Interested in Hindi Technical contents? Follow दी स्मार्ट इनोवेटर