OpenAI's Daybreak project uses Codex Security to automate cyber defense
The new program scans codebases and patches vulnerabilities automatically using advanced AI.
OpenAI officially unveiled Daybreak, a cybersecurity initiative that CEO Sam Altman says is designed to 'accelerate cyber defense and continuously secure software.' Announced on May 11, 2026, Daybreak directly competes with Anthropic's Project Glasswing, which launched the previous month using the Claude Mythos Preview model. Unlike Anthropic's secretive and ominous rollout—where the model was deemed too dangerous for general release—OpenAI's Daybreak takes a transparent, customer-friendly approach. The public landing page features clear calls-to-action: 'Request a vulnerability scan' and 'Contact sales,' with a simple form for interested parties.
Daybreak is powered by Codex Security, a research preview OpenAI introduced in March 2026. The system first builds a 'threat model' of a target application—mapping functions, trust boundaries, and potential vulnerabilities. It then dives into the actual codebase to detect real-world exploits and, in theory, automatically patches them. This AI-driven approach promises continuous, scalable security for corporate and government partners. By making the program open and accessible, OpenAI signals a shift toward democratizing advanced cyber defense, contrasting sharply with Anthropic's restricted, VIP-only model.
- Daybreak uses Codex Security to generate threat models and automatically patch vulnerabilities in codebases.
- OpenAI positions Daybreak as a transparent alternative to Anthropic's Project Glasswing, with a public request form rather than secretive partnerships.
- The initiative follows a March 2026 research preview of Codex Security and aims to secure software continuously for a wide range of companies.
Why It Matters
OpenAI's Daybreak could make AI-driven cybersecurity mainstream, offering proactive defense without the exclusivity of Anthropic's approach.