OpenAI Launches a Specialized Model for Cybersecurity

OpenAI has launched GPT-5.6-Cyber — a specialized vulnerability-research model available exclusively to authorized cybersecurity professionals. It handles 95% of sensitive security queries that standard GPT-5.6 Sol completes in only 1.5% of cases. The previous generation of the same model reached 57.3%, making the jump dramatic.

1Vulnerabilities Found in Real-World Systems

In live testing, GPT-5.6-Cyber discovered two previously unknown vulnerabilities in Chrome's V8 engine that can be chained together to cause memory corruption. In a widely used mobile operating system, the model identified at least five flaws, including a privilege-escalation chain that grants full administrative control over a device.

2Restricted Access with Multi-Layer Verification

The model is available only through the Daybreak Red program: identity verification is required, hardware security keys have been mandatory since September 1, 2026, along with legal declarations and continuous monitoring. OpenAI rated GPT-5.6-Cyber as "High" risk under its Preparedness Framework — not "Critical," but with a note that the trajectory toward the critical threshold steepens with each generation. The same technology in an attacker's hands would radically change the speed and scale of cyber threats.