Daybreak

The defense the AI era demands.

Explore Daybreak solutions Contact Cyber sales

Building the future of cyber defense

OpenAI Daybreak brings together frontier cyber models, Codex Security, trusted workflows, and ecosystem partnerships to help defenders keep pace with an accelerating threat landscape: finding, validating, and fixing vulnerabilities before attackers can exploit them.

The bottleneck in cybersecurity is shifting. AI can now help uncover more security issues across large, complex codebases, but reports alone do not make systems safer. Real protection comes from validated findings, tested patches, coordinated disclosure, maintainer review, and fixes that actually land.

Daybreak is built to accelerate that full remediation loop, working with the world’s leading cyber organizations as partners to bring trusted defensive capability into the tools, services, and workflows security teams already rely on. Through Codex Security, Patch the Planet⁠(opens in a new window), Daybreak models, and the Daybreak Cyber Partner Program, developers, maintainers, researchers, enterprises, and public institutions can turn frontier AI capability into measurable risk reduction.

This work has to happen responsibly. Daybreak is designed around authorization, human judgment, monitoring, safeguards, and collaboration with the broader security community. Advanced access is available for verified defenders through Daybreak Access, pairing more capable and permissive defensive tools with stronger verification, scope controls, and oversight.

Cybersecurity is entering a new era

Frontier models can now perform longer, more complex sequences of cyber work. That creates new opportunities for defenders, and new responsibilities for how these systems are deployed.

Frontier performance

GPT‑5.6 Sol completed the 32-step “The Last Ones” simulation in 7 of 10 attempts, compared with 2 of 10 for GPT‑5.5

Security specialization

Daybreak models bring frontier capabilities to broad defensive workflows. Daybreak Red is specialized for advanced, authorized vulnerability research, exploit validation, penetration testing, and red teaming.

Real-world impact

OpenAI researchers used Daybreak Red to identify two previously unknown vulnerabilities in V8 that could be chained to escape the heap sandbox. Google fixed the first; the second remains under coordinated disclosure.

Explore Daybreak solutions

Built with the world’s best defenders

Daybreak partners with leading security companies, systems integrators, and specialist firms to bring frontier AI into the tools and workflows defenders already trust.

Become a Daybreak partner

Work with OpenAI technical teams to combine frontier cyber capabilities with your expertise, products, and customer relationships—and build differentiated security offerings.

Become a Daybreak partner

Work with a Daybreak partner

Bring frontier AI cyber capability into the tools and services your organization already trusts, with expert implementation, clear safeguards, and accountable delivery.

Explore Daybreak Partners

Fixing the software the world runs on

Open source underpins much of the internet and an estimated 70–90% of modern software⁠(opens in a new window). But many critical projects are maintained by small teams with limited time and funding.

OpenAI wants the people maintaining this shared infrastructure to benefit directly from frontier AI—not be overwhelmed by more unvalidated reports.

Patch the Planet, built with Trail of Bits, pairs frontier models with expert review to validate findings, develop and test patches, coordinate disclosure, and keep maintainers in control of what lands.