AI tools and news, explained in plain words
Models

GPT-6 Cyber: what it is and who can use it

OpenAI's GPT-6 Cyber finds and tests software flaws, with 948 Linux exploits and 81 Firefox bugs so far. Who can get access and what it means for you.

Checked and updated 12 November 2026 at 21:20 UTC

OpenAI's announcement art for GPT-6 Cyber
Image: OpenAI

Key points

  • GPT-6 Cyber is an OpenAI model built to find, prove and help patch security flaws, released on November 5, 2026.
  • It is not in ChatGPT. Only approved Daybreak Red organizations with a separate model approval can use it, through the API and Codex.
  • OpenAI says it built working exploits for 948 Linux kernel bugs and found 81 new high-severity Firefox flaws that Mozilla is fixing.

GPT-6 Cyber is OpenAI's specialist model for security research: it hunts for software vulnerabilities, proves they can be exploited, and helps write fixes. You cannot use it in ChatGPT; OpenAI gives it only to a small group of approved security teams through the API and Codex.

It launched on November 5, 2026, and it is back in the news this week because of OpenAI's claim that it has already uncovered and validated thousands of exploits in widely used code.

What OpenAI says it found

These are OpenAI's own numbers from its launch post:

Target Result
Linux kernel Reproduced 1,724 known vulnerabilities and built working exploits for 948 of them, versus 166 for GPT-5.6 Sol with the same setup. 304 critical findings were sent to maintainers with draft patches.
Firefox 81 previously unknown flaws that Mozilla rated high severity, plus 30 known issues shown to be worse than thought. Mozilla has shipped fixes for many of them.
Brotli (compression library used by Node.js) Helped show how a critical flaw could allow remote code execution in apps that use it.

Security teams at companies including Mozilla, Cisco, Palo Alto Networks, CrowdStrike and CodeRabbit appear in OpenAI's customer list.

How it scores

  • ExploitGym (can the model exploit a known flaw on a remote target): 48.5%, up from 32.1% for GPT-5.6 Cyber.
  • Web penetration testing (find flaws without source code): 40.3% of vulnerabilities found, against 32.0% for the general GPT-6 Astra model.
  • Staying in scope: in a test full of tempting off-limits services, it left them alone 99.93% of the time without its safety prompt and 100% with it.

OpenAI rates the model Critical for cybersecurity under its Preparedness Framework, one step up from High for GPT-5.6 Cyber.

Who can use it

Access runs through Daybreak, OpenAI's trusted access program for security work. There are three levels:

Level What you get
Standard Normal ChatGPT and API models with normal safety rules. No cyber models.
Daybreak Blue Fewer refusals on regular models for verified defensive work, such as code review and malware analysis.
Daybreak Red GPT-5.6 Cyber, and GPT-6 Cyber only with a separate approval, for authorized penetration tests and exploit work.

Individuals (18 or older) and organizations can apply. OpenAI reviews every request, asks for identity and security checks, and requires things like hardware security keys for individuals and IP allowlisting for Red API access. OpenAI says it will widen GPT-6 Cyber to all Daybreak Red customers "soon" but gives no date. The launch post does not list a price.

What it means for normal users

You will not chat with GPT-6 Cyber, but you may benefit from it. The Firefox and Linux bugs it found are being patched, so keeping your browser, phone and computer updated is how those fixes reach you.

It is also a reminder of why companies are cautious about AI agents that act on their own: OpenAI pairs this model with monitoring, blocks on attacks against power and water systems, and a $1 billion grant program to give defenders access first. If you want AI tools you can actually use today at no cost, see our list of free AI tools.

AI Tools This Week is not affiliated with OpenAI.

Frequently asked questions

Can I use GPT-6 Cyber in ChatGPT?

No. OpenAI offers it only through the API and Codex to a select group of approved Daybreak Red customers, and Daybreak Red itself needs a separate approval with stronger identity checks.

What is OpenAI Daybreak?

Daybreak is OpenAI's trusted access program for security work. Daybreak Blue lowers refusals on regular models for verified defenders. Daybreak Red adds the cyber models for authorized penetration testing and exploit work. Individuals 18 and over and organizations can apply, and approval is not automatic.

Is GPT-6 Cyber dangerous?

OpenAI rates its cyber capability as Critical under its Preparedness Framework, the highest level so far for a cyber model. That is why access is limited and paired with monitoring, IP allowlisting and blocks on attacks against critical infrastructure.

Sources

  1. GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents (OpenAI, November 5, 2026)
  2. OpenAI Daybreak: Trusted Access for Cyber overview (OpenAI Help Center)

The AI Tools This Week Desk writes and edits every article, checks claims against primary sources and updates stories when the facts change.

Related

Mentioned in