OpenAI shares Defense Factory playbook: 250+ staff, AI agents hunting vulnerabilities
gdb · x · 2026-09-10
OpenAI president Greg Brockman shared the architecture and a practical playbook for how the company secures its own systems. OpenAI mobilized 250+ people across hundreds of systems, using its latest cyber models to find and fix vulnerabilities that might never have been discovered otherwise. The published "Defense Factory" is a continuous loop where AI agents find vulnerabilities, validate them, and verify fixes — offered so other organizations can make the most of the defenders' window.
More from Safety
- Reasoning models can barely control their chain of thought, paper finds — tomekkorbak · 2026-10-27
- LessWrong: GLM models show outlier chain-of-thought controllability vs Kimi and DeepSeek — tomekkorbak · 2026-10-27
- Claude's refusal to type one-time passwords is a real security impediment — AaronBergman18 · 2026-10-27
- First systematic data center noise study: 17 of 25 hyperscalers drew no credible complaints — AndyMasley · 2026-10-27
- UK AISI: under $80 of fine-tuning fully restores filtered bio capabilities in open-weight models — connoraxiotes · 2026-10-27
- Meta joins frontier labs reporting CoT controllability evals in Muse Spark 1.3 card — tomekkorbak · 2026-10-27