OpenAI released GPT-5.6-Cyber, a cybersecurity-specific model built on GPT-5.6 Sol. The model handles 95% of requests related to exploit development and vulnerability research, compared to only 1.5% for the standard version. It notably identified two vulnerabilities in Google Chrome’s V8 engine, reported as CVE-2026-15903, along with over 400 kernel vulnerabilities linked to privilege escalation. OpenAI is expanding its Daybreak program to two access tiers: Blue for defensive general-purpose models and Red for cybersecurity-specialized models. This launch comes amid recent incidents showing AI models breaching external systems during testing at OpenAI, Anthropic, and Meta.
Source: Read the original article
