GPT-6 Cyber
OpenAI released GPT-6 Cyber, a security model rated Critical under its Preparedness Framework, scoring 48.5% on ExploitGym, to verified Daybreak Red users first.
GPT-6 Cyber improves on GPT-5.6 Cyber, which was rated High, and beats GPT-6 Astra on most cyber tasks. OpenAI says it follows task scope more reliably, avoiding out-of-scope exploitation in 99.93% of alignment tests without its deployment prompt. Researchers used it to build working exploits for 948 Linux kernel vulnerabilities and to find 81 previously unknown high-severity Firefox flaws.
- Date
- Thursday 5 November 2026
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| ExploitGym score | 48.5% up from 32.1% for GPT-5.6 Cyber, a 51% relative gain | company |
| Web penetration testing recall | 40.3% versus 32.0% for GPT-6 Astra, on a new OpenAI evaluation | company |
| RemoteExploitBench tasks completed | 95% GPT-6 Astra with Daybreak Blue refused 100% of tasks | company |
| Linux kernel exploits developed | 948 of 1,724 reproduced vulnerabilities privilege escalation exploits, about $1.2 million in API credits | company |
| Firefox high-severity vulnerabilities found | 81 previously unknown; Mozilla has released fixes for many | company |
Access is limited to verified customers in the Daybreak Red program, with IP allowlisting required by default. Daybreak is being expanded to more organizations and a new tier for critical infrastructure operators.
Sources
- openai.com/index/introducing-gpt-6-cyber
- axios.com/2026/11/05/openai-linux-cyber-model
- thehackernews.com/2026/11/openai-launches-gpt-6-cyber-after-ai.html
- cxodigitalpulse.com/openai-releases-gpt-6-cyber-under-restricted-security-access/
This record was checked and corrected against its sources on 5 November 2026. How we check