On 5 November 2026, OpenAI announced GPT-6 Cyber, a new version of its GPT-6 model built for one job: finding and fixing software vulnerabilities. The company says the model is better at discovering security flaws, building working exploits from them, and staying inside the limits researchers set for it. To show what it can do, OpenAI published results from its own testing: the model found 81 previously unknown high severity vulnerabilities in Firefox and developed working exploits for 948 vulnerabilities in the Linux kernel.
The launch matters because it is the first time OpenAI has released a model aimed squarely at cybersecurity work, and it comes with its highest risk rating: Critical for cybersecurity under its Preparedness Framework, up from High for the earlier GPT-5.6 Cyber.
What Is GPT-6 Cyber?
GPT-6 Cyber is a specialised version of OpenAI’s GPT-6 model family, tuned for authorised security research. It is designed to search software for vulnerabilities, develop working exploits that prove a flaw is real, and propose patches to fix it. OpenAI describes it as a major step forward for security research and for defenders, because it combines stronger vulnerability detection with better alignment, meaning it is more reliable at staying within the boundaries researchers set.
The model is aimed at professional security teams, not general users. Its intended uses are vulnerability research, exploit development for testing, and authorised penetration testing, where a tester has permission to probe a system for weaknesses.
What GPT-6 Cyber Found in Firefox and Linux
OpenAI tested the model on two widely used pieces of software and published the results with the announcement. In Firefox, GPT-6 Cyber produced evidence for 81 previously unknown vulnerabilities that Mozilla classified as high severity, along with proposed patches. The model also showed that 30 known issues were worse than first reported. Mozilla has confirmed it is working with frontier AI models from OpenAI and Anthropic to speed up Firefox security fixes. The testing also found a critical vulnerability in Brotli, the compression library, tracked as CVE-2026-111497 with a CVSS score of 9.2.
In the Linux kernel, OpenAI ran the model through an automated research pipeline against the latest kernel development code. The model reproduced 1,724 vulnerabilities and developed working local privilege escalation exploits for 948 of them. A local privilege escalation exploit lets an unprivileged local user obtain root access in a test environment, which makes it a serious class of flaw. For comparison, GPT-5.6 Sol managed working exploits for 166 vulnerabilities using the same research setup, so GPT-6 Cyber produced more than five times as many.
How It Scores on Security Benchmarks
OpenAI also reported benchmark results. On ExploitGym, a benchmark that tests whether an AI agent can turn known software vulnerabilities into functioning real world exploits, GPT-6 Cyber scored 48.5 percent, compared with 32.1 percent for GPT-5.6 Cyber. OpenAI describes that as a 51 percent relative improvement. The model also outperformed GPT-6 Astra on most cybersecurity tasks in OpenAI’s testing.
| Test | GPT-6 Cyber | Previous model |
|---|---|---|
| ExploitGym score | 48.5% | 32.1% (GPT-5.6 Cyber) |
| Linux kernel exploits built | 948 | 166 (GPT-5.6 Sol, same setup) |
| Linux vulnerabilities reproduced | 1,724 | Not reported for earlier models |
| Firefox flaws found | 81 new high severity flaws | Not reported for earlier models |
Benchmarks like these come from OpenAI’s own testing, so they should be read as the company’s claims about its model. What makes the Firefox and Linux numbers more concrete is that they describe real flaws found in real software, with Mozilla confirming the Firefox work.
Why OpenAI Rated It Critical
Under OpenAI’s Preparedness Framework, models are rated on how much their capabilities could help with harmful uses. GPT-6 Cyber is the first model OpenAI has rated Critical for cybersecurity, a step above the High rating of GPT-5.6 Cyber. The company says the rating reflects the model’s skill at finding vulnerabilities and building exploits, and that it has added safeguards for industrial controls, essential services, election systems and other high risk targets. The rating is also why access is limited. A model that can find flaws at this scale in the wrong hands could help attackers, so OpenAI is releasing it through a controlled programme rather than to the general public.
Who Can Use GPT-6 Cyber
GPT-6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers. Daybreak Red is OpenAI’s programme for trusted security researchers and defenders. Access requires identity checks, organisational security requirements, single sign on, multi factor authentication and API IP allowlisting. OpenAI says wider access for Daybreak Red customers will follow in the coming weeks. Alongside the model, OpenAI announced a 1 billion dollar Daybreak grant initiative for frontline cybersecurity defenders and organisations with limited resources, including government bodies and essential service providers.
What It Means for Users
For most people, GPT-6 Cyber will not change anything directly. You cannot sign up for it as a consumer product, and it is not a tool you will use yourself. Its effect shows up indirectly: flaws in Firefox, Linux and other software get found and patched faster, before attackers can use them. For security teams, the launch is a sign of where AI assisted defence is heading. Models that can find flaws, prove them with working exploits and suggest patches could shorten the time between a vulnerability appearing and a fix reaching users. The risk is the same capability in the wrong hands, which is why OpenAI is keeping access controlled and why the Critical rating matters.
For background on staying safe online, read our guides on the best VPNs in 2026 and the DigiCert data breach, and see how companies respond when they are hit, in our report on the DreamHost data breach. For the wider reporting, see The Hacker News, Cyber Magazine and Security Affairs.
Frequently Asked Questions
What is GPT-6 Cyber?
It is a specialised version of OpenAI’s GPT-6 model built for authorised cybersecurity research: finding software vulnerabilities, developing working exploits to prove them, and proposing patches.
When was GPT-6 Cyber announced?
OpenAI announced GPT-6 Cyber on 5 November 2026.
What did GPT-6 Cyber find?
In OpenAI’s testing it found 81 previously unknown high severity vulnerabilities in Firefox, reproduced 1,724 Linux kernel vulnerabilities and built working exploits for 948 of them, and found a critical Brotli vulnerability tracked as CVE-2026-111497.
How good is it compared with earlier models?
On the ExploitGym benchmark it scored 48.5 percent, against 32.1 percent for GPT-5.6 Cyber. It built 948 Linux kernel exploits where GPT-5.6 Sol managed 166 with the same setup.
Can anyone use GPT-6 Cyber?
No. It is initially available through the API and Codex only to a select group of approved Daybreak Red customers, with identity checks, single sign on, multi factor authentication and API IP allowlisting. Wider access is promised in the coming weeks.
Why is GPT-6 Cyber rated Critical?
Under OpenAI’s Preparedness Framework it is the first model rated Critical for cybersecurity, up from High for GPT-5.6 Cyber. The rating reflects its ability to find vulnerabilities and build exploits, and OpenAI says it added safeguards for high risk targets such as industrial controls, essential services and election systems.
#GPT6Cyber #OpenAI #CyberSecurity #VulnerabilityResearch #Firefox #Linux