OpenAI has launched Daybreak, a new cybersecurity initiative aimed at competing with Anthropic’s Project Glasswing. Anthropic’s Project Glasswing leverages the unreleased AI model, Claude Mythos Preview, to enhance cyber defense for its clients. Mozilla reported in April that the Claude Mythos model helped identify and patch 271 vulnerabilities in the Firefox browser.
Daybreak utilizes multiple OpenAI AI models, including the specialized security agent Codex. OpenAI stated that the initiative is designed around the principle that cybersecurity should be integrated into software development from the outset, rather than focusing solely on remediation after vulnerabilities are identified.
The initiative seeks to prioritize significant cybersecurity issues and decrease analysis times from hours to minutes. Daybreak is capable of generating and testing patches within code repositories and providing audit-ready evidence for clients’ systems. OpenAI demonstrated Daybreak’s capabilities by having Codex Security scan a codebase, validate critical findings, and implement necessary fixes.
Daybreak will employ GPT-5.5 for general purposes and for most defensive workflows, including a version of GPT-5.5 with Trusted Access for cybersecurity functions. Key workflows for Daybreak include secure code reviews, vulnerability triage, malware analysis, detection engineering, and patch validation. Additionally, GPT-5.5-Cyber will be used for specialized tasks such as authorized red teaming, penetration testing, and controlled validations.
OpenAI is collaborating with several partners for the Daybreak initiative, which includes companies like Cloudflare, Cisco, CloudStrike, Palo Alto Networks, Oracle, and Akamai.
