OpenAI's New AI Model Helps Approved Security Researchers Find Software Flaws

OpenAI has released GPT-5.6-Cyber, a specialized AI model available through its Daybreak Red cybersecurity program. The model is trained to help with sensitive security tasks — like finding undiscovered software vulnerabilities and developing methods to exploit them — that ordinary AI assistants are programmed to refuse. Access is restricted to approved partners doing authorized security work.