Recent events

OpenAI launches GPT-5.6-Cyber cybersecurity AI model

OpenAI launched GPT-5.6-Cyber, a cybersecurity-focused AI model fine-tuned from the flagship GPT-5.6 Sol, aimed at vulnerability research, penetration testing and incident response. It is not available to general users and is provided only to vetted enterprises and security vendors through the Daybreak program. The model achieves a 95% completion rate on high-risk tasks, up from 57.3% for GPT-5.5-Cyber, and discovered unknown vulnerabilities in the Chrome V8 engine and a mobile operating system. The Daybreak program has two tracks, Blue for defensive teams and Red for authorized researchers, with initial partners including Accenture, IBM and Palo Alto Networks. Hardware security keys will be mandatory from September.