Build responsible AI applications with Guardrails for Amazon Bedrock
summary
Configures Bedrock guardrails that intercept both user prompts and model outputs to enforce denied topics (e.g., investment advice, cryptocurrency), six content categories including hate and sexual violence, and prompt attack filter for injection and jailbreak attempts. Demonstrates agent integration using Anthropic Claude Haiku where investment and combined bitcoin+insult inputs are flagged and replaced with a custom blocked message, and a summarization pipeline where PII masking redacts emails and booking IDs via custom regex, all model‑agnostic and working in the console test traces.
AI-written summary · about
chapters (6)
- 0:00Introduction to Gen AI safety
- 2:48Guardrails for Bedrock
- 3:58Guardrail policy types
- 7:58Guardrail configuration demo
- 12:19Agents and security
- 15:26Redacting sensitive data
related talks
41mBuild AI applications with Guardrails for Amazon Bedrock
20mResponsible AI with Amazon Bedrock Guardrails
39mBuilding Safe AI Agents
26mSponsored Session: Navigating Open Source AI with Guardrails
31mUsing Guardrails.ai for building Generative AI apps
26mOne guardrail is not enough: Defense-in-depth against LLM jailbreaks
30mWhy Prompt-Level Guardrails Fail for AI Agents
30mAccelerate AI to prod