[UPDATE] OpenAI's GPT-5.6-Cyber ships to approved defenders through the listed security vendors
OpenAI released GPT-5.6-Cyber on 2026-08-10, expanding its Daybreak security programme into two tiers, and assessed both GPT-5.6 Sol and the Cyber variant as reaching the 'High' cybersecurity threshold under its Preparedness Framework — neither at 'Critical'. Vendor testing reported a 95% completion rate on exploit-chain, privilege-escalation and authentication-bypass prompts against 1.5% for GPT-5.6 Sol, and the model is described as having surfaced previously unknown vulnerabilities in Chrome's V8 engine. Access is restricted to selected firms and to supported security vendors named as Palo Alto Networks, CrowdStrike, Cisco, Sophos, Akamai, Fortinet and Cloudflare. That distribution structure positions the listed vendors as channels for frontier-lab cyber capability rather than as its displacement targets.