OpenAI removes GPT-5.6 safety guardrails for exploit research — the Daybreak program goes operational
On August 10, 2026, OpenAI launched GPT-5.6-Cyber, a variant of its flagship model with significantly reduced safety refusals, designed for offensive cybersecurity teams. The Daybreak program, initially a research partnership, is now an operational tool. Here is what this means for red teams and defenders.