OpenAI Unfiltered Models for Cybersecurity Professionals

OpenAI expanded the Daybreak program, splitting it into two tiers: a defensive Blue tier based on GPT-5.6 Sol without filters, and an offensive Red tier based on the new GPT-5.6-Cyber model, which demonstrated a 95% success rate on cyber tasks compared to 1.5% for the standard model.

Compiled by Sergey KostenchukPublished 2026-08-11Updated 2026-09-18

2026-08-11 Safety OpenAI

🛡 OpenAI Unfiltered Models for Cybersecurity Professionals

OpenAI split the Daybreak program into two tiers: Blue (GPT-5.6 Sol without filters for defense) and Red — the new GPT-5.6-Cyber model for penetration testing. It handles 95% of cyber requests compared to 1.5% for a standard model, discovered vulnerability CVE-2026-15903 in Chrome, 400+ vulnerabilities in the OS kernel, and 3 critical bugs in a database management system.

🌍 A direct response to Anthropic's Project Glasswing. Starting September 1, 2026, hardware security keys will be mandatory.

👤 For cybersecurity professionals — tasks that used to take weeks are now solved in a day. Update Chrome — the CVE-2026-15903 patch is already out.

Source 1: https://cybersecuritynews.com/openai-expands-daybreak-cyber/ Source 2: https://cyberpress.org/openai-expands-daybreak-with-gpt-5-6-cyber/

Sources