AI Safety & Cybersecurity
OpenAI details GPT-6 Cyber for security work
The company says access is restricted to approved users; its system card does not verify claims about Firefox or Linux vulnerabilities.
Conceptual illustration of a cybersecurity researcher examining abstract code patterns on a monitor.
Photo: Kronos News
OpenAI published a system card for GPT-6 Cyber, a cybersecurity-focused version of GPT-6 Astra intended for authorized work such as finding vulnerabilities, developing exploits and testing websites. The company says the model’s specialized training aims to improve these capabilities while keeping it within users’ authorized scope. [1]
Access is restricted through OpenAI’s Daybreak Red program, with identity checks and organizational security requirements, including single sign-on and multi-factor authentication. OpenAI also describes safeguards intended to block high-risk assistance involving critical systems and to monitor whether activity matches a user’s authorization. [1]
In an external evaluation described in the system card, GPT-6 Cyber solved 90 of 288 challenges, compared with 96 for GPT-6 Astra; evaluators reported no successful attacks on fully hardened targets, and neither model solved the 15 hardest challenges. The supplied source does not substantiate the brief’s claims about 81 Firefox vulnerabilities or 948 Linux kernel exploits, so those figures are not included as verified findings. [1]
Editorial notes
Transparency note
AI assisted drafting. Human edited and reviewed.
- AI assisted
- Yes
- Human review
- Yes
- Last updated
Risk assessment
OpenAI’s system card is the sole supplied source and does not verify the specific Firefox and Linux figures in the editorial brief.
Sources
Related stories
View all
Topics
About the author
Kronos News Desk covers ai safety & cybersecurity and editorial analysis for Kronos News.