OpenAI launches GPT-6 Cyber for vetted security researchers

New cyber-specialized model reaches vetted Daybreak Red users, with higher benchmark scores and tighter access controls.

In this brief: 2 sections 1 min read

- OpenAI reports 48.5% on ExploitGym versus 32.1% for GPT-5.6 Cyber. It rates GPT-6 Cyber Critical for cybersecurity. Initial access is restricted to select Daybreak Red customers.
- Approved security teams and researchers can assess Daybreak Red eligibility, system-card limits and authorization controls. No fixed application deadline was announced.

Useful?