OpenAI has released Daybreak, a new cybersecurity platform that combines its frontier models with Codex (an agent runtime framework) to allow security teams to generate, test, and validate security patches directly in their codebases.
To balance power and potential misuse, Daybreak implements a strict three-tier access system:
- Regular GPT-5.5 with standard safety guardrails
- GPT-5.5 with Trusted Access, which features more precise guardrails designed for everyday defense tasks like malware analysis and vulnerability triage
- A highest-permission tier, intended for penetration testing and other high-stakes security assessments