OpenAI launches Daybreak cybersecurity initiative to rival Anthropic's Claude Mythos

OpenAI has officially launched Daybreak, a new cybersecurity initiative powered by GPT-5.5 and Codex Security. The move positions the initiative as a direct competitor to Anthropic's Project Glasswing, which relies on the unreleased Claude Mythos model. While Anthropic restricted access to its high-capability model due to safety concerns, OpenAI is deploying Daybreak publicly to integrate cyber defence directly into the software development lifecycle.

The core strategy behind Daybreak is to embed security measures from the outset of development rather than relying solely on post-hoc vulnerability fixing. The system is designed to prioritise high-impact issues and reduce the time required for analysis from hours to minutes. Furthermore, it aims to generate and test patches within code repositories, sending results back to client systems with audit-ready evidence.

Technically, the initiative utilises GPT-5.5 for general purposes and GPT-5.5 with Trusted Access for Cyber for most defensive workflows. These capabilities include secure code review, vulnerability triage, malware analysis, detection engineering, and patch validation. A specialised variant, GPT-5.5-Cyber, is reserved for preview access to high-risk workflows such as authorised red teaming, penetration testing, and controlled validation.

This launch follows the public release of GPT-5.5 capabilities and fills a gap in OpenAI's portfolio. It contrasts with Anthropic's approach, which previously demonstrated efficacy through Mozilla reporting that Mythos helped identify and patch 271 vulnerabilities in the Firefox browser in April. OpenAI emphasises that its strategy focuses on building defence into software from the start, a shift prompted by the public release of its latest models.

To support this effort, OpenAI has announced collaborations with major security firms including Cloudflare, Cisco, CloudStrike, Palo Alto Networks, Oracle, and Akamai. These partnerships are intended to bolster the initiative's ability to validate the highest-risk findings and fix them within the codebase.

While the launch signals a new era of automated security, specific efficacy metrics comparing Daybreak to existing manual or automated tools have not yet been publicly disclosed. There is also no independent verification provided in the source material regarding the actual performance of Daybreak in real-world scenarios versus the claimed reduction in analysis time.