OpenAI hands vetted security firms a model that answers 95% of hacking prompts

OpenAI split Daybreak into Blue and Red tiers and released GPT-5.6-Cyber, a hacking-focused model that answers 95% of advanced cyber prompts.