Media & Culture

OpenAI's Daybreak uses GPT-5.5-Cyber to hunt security flaws

Daybreak automates threat modeling and patches vulnerabilities before attackers strike

Deep Dive

OpenAI has unveiled Daybreak, a security-focused AI initiative that leverages the company's most advanced models—including GPT-5.5-Cyber and GPT-5.5 with Trusted Access for Cyber—alongside the Codex Security agent (launched in March). Daybreak automatically builds a threat model based on an organization's codebase, identifies possible attack paths, validates which vulnerabilities are realistic, and prioritizes the highest-risk ones for automated patching. The system is designed to stay ahead of attackers by finding and fixing flaws before they can be exploited.

Daybreak arrives just over a month after Anthropic announced Claude Mythos, a security AI it claimed was too dangerous to release publicly and instead shared only privately via Project Glasswing—though some unauthorized parties still gained access. OpenAI says Daybreak is not a single model but a collaborative framework involving security partners and government agencies. The company plans to "deploy increasingly more cyber-capable models" as the initiative evolves, positioning Daybreak as a proactive defense alternative to Anthropic's more restricted approach.

Key Points
  • Daybreak uses GPT-5.5-Cyber and the Codex Security agent to automate vulnerability detection and patching
  • It competes with Anthropic's Claude Mythos, which was deemed too dangerous for public release
  • OpenAI says Daybreak involves industry and government partners to enable future cyber-capable models

Why It Matters

Enterprises can now automate proactive security patching, shifting from reactive defense to AI-driven threat prevention.