用合成資料測試 Amazon Bedrock Guardrails
AI guardrails can block useful questions and still let unwanted requests through. This talk shows how to use synthetic data to find those gaps in Amazon Bedrock Guardrails.
Using a financial assistant as an example, you’ll learn how to turn business rules into realistic test questions, check the quality of AI-generated test data, and see whether a settings change fixes one problem while creating another. The goal is to help you test guardrails with confidence, rather than assume they work because they’re switched on.
LLM 接上護欄(Guardrails)後怎麼知道它有沒有生效?本場用 Amazon Bedrock 合成出測試資料、用於驗證 Amazon Bedrock Guardrails;並以金融資訊助理為例驗證一般金融問題能不能正常回答、而超過任務範圍的請求會不會被攔下。
也會分享如何檢查測資、找出誤擋與漏擋,再調整及複測護欄是否真的有效。
Nero Un
November 04, 2026