OpenAI Mobilizes 250+ Staff to Build a "Defense Factory" Where AI Agents Hunt and Patch Vulnerabilities
OpenAI announced a company-wide code-red security sprint mobilizing 250+ people to harden hundreds of internal systems. Its latest cybersecurity models and Codex agents wrote all the patches, discovering and fixing critical bugs that humans might never have found. OpenAI also published the full "Defense Factory" reference architecture, case study, and playbook for other organizations to replicate. The news came from OpenAI's official channels and was shared by co-founder Greg Brockman.
Confirmed
- OpenAI mobilized 250+ staff across hundreds of systems for defensive hardening
- Codex agents and their cyber models wrote the patches, finding and fixing many otherwise-undiscoverable critical bugs
- OpenAI published the Defense Factory reference architecture, case study, and playbook for others to replicate
- According to Meta researcher Martin Signoux, OpenAI's Daybreak project is helping European cyber defenders find and fix vulnerabilities
Why it matters
- A large-scale practice of systematically applying AI agents to a company's own security, demonstrating real value in the vulnerability discovery-validation-patching loop
- Publishing the architecture and playbook lets other enterprises adopt the process, potentially upgrading industry-wide security operations
- Daybreak extends the approach to external defenders in Europe, moving AI security from internal practice toward industry collaboration
2026-09-10 ~ 2026-09-11 · 8 related posts
Primary sources
- OpenAI Mobilized 250+ People to Build a "Defense Factory" of AI Agents That Find and Fix Vulnerabilities — OpenAI ·
- OpenAI used its own models to find and fix critical vulns in a 250+ person defense effort — gdb ·
- OpenAI's Daybreak helps European cyber defenders find and fix vulnerabilities — MartinSignoux ·
- [source] OpenAI Mobilized 250+ People to Build a "Defense Factory" of AI Agents That Find and Fix Vulnerabilities — OpenAI · 2026-09-10
- OpenAI used its own models to find and fix critical vulnerabilities in 250+ person security effort — zetalyrae · 2026-09-10
- OpenAI's 250-person security sprint: Codex agents wrote every patch across hundreds of systems — rohanpaul_ai · 2026-09-10
- [source] OpenAI's Daybreak helps European cyber defenders find and fix vulnerabilities — MartinSignoux · 2026-09-11
4 near-duplicate retellings: gdb · gdb · AdtRaghunathan · MartinSignoux