OpenAI ships GPT-6 Cyber, but only to approved Daybreak Red customers

OpenAI released GPT-6 Cyber on November 5: a security model built to find vulnerabilities and develop working exploits while staying inside the scope a researcher sets. It is not public. OpenAI is giving it, through the API and Codex, to a select group of approved Daybreak Red customers first. ArtificialWatch's radar saw a GPT-6 Cyber id in a catalog sweep at 17:08:44 UTC and fired at 17:09:46 UTC.

OPENAI'S OWN FIGURES

- ExploitGym: 48.5%, against 32.1% for GPT-5.6 Cyber.

- RemoteExploitBench, a new OpenAI test: 95% of tasks completed, where GPT-6 Astra on Daybreak Blue refused all of them.

- A new web-pentesting test: 40.3% recall, against 32.0% for GPT-6 Astra.

- OpenAI rates it Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber.

WHAT IT HAS FOUND, PER OPENAI

Reproducible evidence for 81 novel Firefox vulnerabilities that Mozilla classed as high severity, and working exploits for 948 of the 1,724 Linux kernel vulnerabilities it reproduced, against 166 for GPT-5.6 Sol on the same harness. That makes it the unnamed model behind OpenAI's November 4 Linux kernel post.

STAYING IN SCOPE

On an ExploitGym honeypot test, it left out-of-scope services alone in 99.93% of runs without its deployment prompt and in all of them with it. GPT-5.6 Cyber managed 94.49%.

WHO CAN USE IT

Approved Daybreak Red customers, through the API and Codex. OpenAI says all Daybreak Red customers get it soon, and opens GPT-5.6 Cyber to all of them now. Daybreak Red API access requires IP allowlisting by default. OpenAI published no price.

WHAT WE DO NOT KNOW

No independent benchmark has scored it, and OpenAI gave no date for wider access. When OpenAI first named it on October 23, it called it GPT-6 Astra Cyber.