OpenAI has launched Daybreak, a cybersecurity initiative aimed at competing with Anthropic’s Project Glasswing. Project Glasswing utilizes Anthropic’s unreleased AI model, Claude Mythos Preview, which helped Mozilla find and patch 271 vulnerabilities in Firefox, according to a report from April.
Daybreak employs various AI models from OpenAI, including the Codex security agent. The initiative is premised on integrating cyber defense into software development from the outset, rather than focusing only on identifying and fixing vulnerabilities.
OpenAI stated that Daybreak seeks to prioritize high-impact issues and reduce analysis time from hours to minutes. It will generate and test patches within repositories and furnish clients with audit-ready evidence. In a demonstration, OpenAI asked Codex Security to scan a codebase, validate high-risk findings, and implement fixes.
The initiative will utilize GPT-5.5 for general purposes and GPT-5.5 with Trusted Access for Cyber for defensive security workflows. These workflows encompass secure code review, vulnerability triage, malware analysis, detection engineering, and patch validation. Additionally, GPT-5.5-Cyber will support specialized workflows, including authorized red teaming and penetration testing.
OpenAI is collaborating with several partners for Daybreak, including Cloudflare, Cisco, CloudStrike, Palo Alto Networks, Oracle, and Akamai.
