Membina masa hadapan pertahanan siber
OpenAI Daybreak menghimpunkan model siber perbatasan, Codex Security, aliran kerja yang dipercayai dan perkongsian ekosistem untuk membantu pasukan pertahanan mengikuti perkembangan landskap ancaman yang semakin pesat: mencari, mengesahkan dan membaiki kelemahan sebelum penyerang dapat mengeksploitasinya.
Kesesakan dalam bidang keselamatan siber sedang berubah. AI kini boleh membantu mendedahkan lebih banyak isu keselamatan merentas pangkalan kod yang besar dan kompleks, tetapi laporan semata-mata tidak menjadikan sistem lebih selamat. Perlindungan sebenar datang daripada penemuan yang disahkan, tampung yang diuji, pendedahan terselaras, semakan penyelenggara dan pembaikan yang benar-benar diterapkan.
Daybreak dibina untuk mempercepatkan keseluruhan kitaran pemulihan tersebut, bekerjasama dengan organisasi siber terkemuka dunia sebagai rakan kongsi untuk membawa keupayaan pertahanan yang dipercayai dalam alat, perkhidmatan dan aliran kerja yang sememangnya dipercayai oleh pasukan keselamatan. Melalui Codex Security, Patch the Planet, GPT‑5.5‑Cyber dan Daybreak Cyber Partner Program, pembangun, penyenggara, penyelidik, perusahaan dan institusi awam boleh mengubah keupayaan AI perbatasan menjadi pengurangan risiko yang boleh diukur.
Usaha ini perlu dilaksanakan secara bertanggungjawab. Daybreak direka bentuk berteraskan kebenaran, pertimbangan manusia, pemantauan, langkah perlindungan dan kerjasama dengan komuniti keselamatan yang lebih luas. Akses lanjutan tersedia untuk pasukan pertahanan yang disahkan melalui Trusted Access for Cyber, yang menggabungkan alat pertahanan yang lebih berkeupayaan dan lebih permisif dengan pengesahan, kawalan skop dan pengawasan yang lebih kukuh.
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.
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.
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.
Fixing the software the world runs on
Open source underpins much of the internet and an estimated 70–90% of modern software(dibuka dalam tetingkap baru). 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.
What defenders have achieved
Committed to greater cyber resilience
$17M
API credits and direct support for open-source security and the wider maintainer ecosystem.
Codebases under review
41
Open-source projects receiving AI-assisted research and expert security review.
Issues identified
858
Findings surfaced for validation, prioritization, and coordinated disclosure.
Patches produced
263
Targeted fixes developed and tested before maintainer review.
Patches accepted upstream
143
Fixes accepted by maintainers for inclusion in the projects they steward.
OpenAI’s commitment includes API credits and direct support for work with Trail of Bits, Calif, Linux Arkrites, and the wider maintainer ecosystem.
Terokai sumber tambahan
Follow new evaluations, technical learnings, patches, and ecosystem work as they publish.
