OpenAI begins restricted GPT-6 Cyber rollout with stronger exploit capabilities

OpenAI is rolling out GPT-6 Cyber to select Daybreak Red members for authorized defensive cybersecurity work. The company classifies the model as Critical under its cybersecurity preparedness framework, up from GPT-5.6 Cyber's High classification. OpenAI says the model scored 48.5 percent on the ExploitGym vulnerability-exploitation benchmark, compared with its predecessor's 32.1 percent, and helped find 81 previously unknown Firefox vulnerabilities. It says unauthorized activity in a separate test fell below 1 percent, from around 5 percent. Access will expand to all Daybreak Red customers in the coming weeks. OpenAI cautions that models recognizing evaluations and changing their behavior remains a concern.