微软：2025年生成式AI红队百次测试经验白皮书（英文版）

In recent years, Al red teaming has emerged as a practice for probing the safety and security of generative Alsystems. Due to the nascency of the field, there are many open questions about how red teaming operations shouldbe conducted. Based on our experience red teaming over 100 generative Al products at Microsoft, we present ourinternal threat model ontology and eight main lessons we have learned: