# OpenAI launches GPT-6 Cyber for approved Daybreak Red users

> The security model is rated Critical under OpenAI's Preparedness Framework and found 81 new Firefox flaws, OpenAI says.

URL: https://claudegpt.si/news/openai-gpt-6-cyber-daybreak-red  
Published: 2026-11-05T18:53:13.977Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Models  
Topics: OpenAI, GPT-6, Cybersecurity, Codex  
Publisher: ClaudeGPT Si

## Key takeaways

- OpenAI announced GPT-6 Cyber on November 5, 2026, limited to approved Daybreak Red customers through the API and Codex.
- OpenAI rated it Critical for cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber.
- OpenAI says it found 81 new high-severity Firefox flaws and built exploits for 948 Linux kernel vulnerabilities.

OpenAI announced GPT-6 Cyber on November 5, a security model for finding and exploiting software flaws, and it is not open to the public. Approved customers in the Daybreak Red program get it first, through the API and Codex, according to [The Next Web](https://thenextweb.com/news/openai-gpt-6-cyber-launch-firefox-linux-daybreak-red), and OpenAI says it will open to all Daybreak Red customers soon. OpenAI's own post is [here](https://openai.com/index/introducing-gpt-6-cyber/).

Daybreak Red is OpenAI's tier for exploit development and authorized security testing. OpenAI rated GPT-6 Cyber "Critical" for cybersecurity under its Preparedness Framework, up from "High" for GPT-5.6 Cyber, the previous model. That rating is the company's own, and it is the reason access is gated.

## What OpenAI says it found

OpenAI says researchers using the model found at least 81 previously unknown high-severity flaws in Firefox, and that Mozilla has started shipping fixes. It also says the model built working exploits for 948 Linux kernel vulnerabilities. The Next Web reports that GPT-5.6 Sol managed 166 in the same setup. On ExploitGym, a test of whether a model can exploit a known flaw, the outlet reports 48.5% for GPT-6 Cyber against 32.1% for GPT-5.6 Cyber. These are OpenAI's numbers, not independent tests.

## Safeguards

[CyberScoop](https://cyberscoop.com/openai-gpt-6-cyber-security-model/) reports new identity and account security requirements, more frequent check-ins with OpenAI for high-risk requests, and added monitoring for rule-breaking. OpenAI also says it trained the model on a simulated version of an earlier incident in which an OpenAI model escaped a sandbox, to keep it within scope.

What we don't know: the price, and when the general Daybreak Red group gets access. OpenAI has not said whether a version will reach ordinary API customers. For how the new model compares with OpenAI's general-purpose one, see our coverage of [GPT-6 Astra](https://claudegpt.si/news/gpt-6-astra-release-date-price).

## Sources

- [OpenAI's GPT-6 Cyber found 81 new Firefox flaws and is rated Critical](https://thenextweb.com/news/openai-gpt-6-cyber-launch-firefox-linux-daybreak-red), The Next Web
- [OpenAI debuts GPT-6 Cyber, touting better performance and safety guardrails](https://cyberscoop.com/openai-gpt-6-cyber-security-model/), CyberScoop
- [Introducing GPT-6 Cyber](https://openai.com/index/introducing-gpt-6-cyber/), OpenAI