OpenAI launches cybersecurity model GPT-6 Cyber

OpenAI released GPT-6 Cyber, a cybersecurity-focused model it rates "Critical" for security capability, claiming out-of-bounds actions in under 1% of tests simulating the earlier AI-agent Hugging Face intrusion, versus about 5% for GPT-5.6 Cyber. On the ExploitGym vulnerability-exploitation benchmark, GPT-6 Cyber scored 48.5%, ahead of GPT-5.6 Cyber at 32.1% and GPT-6 Astra at 42.4%. OpenAI said the model helped uncover 81 previously unknown Firefox vulnerabilities and has added safeguards to detect and block misuse. Access is initially limited to select members of the Daybreak Red program, with a rollout to all of its customers expected over the coming weeks. The launch signals OpenAI's push to position security-specific models as a distinct product tier, and the pace of the Daybreak Red expansion will show how quickly that capability reaches enterprise buyers.