Building Safe AI Agents
summary
Shows how to lock down agent behavior using Amazon Bedrock Guardrails to filter inputs and outputs and AgentCore Policy (Cedar) to declare what tools and refunds an agent may execute, with the Cedar analyzer proving policy properties. Describes a neurosymbolic loop where an LLM generates code and an automated reasoning verifier rejects or accepts until a provable result is reached. Genesys reports a Genesis Cloud Co‑pilot serving 1,000 tools across 16 regions and 15 languages with <5s latency and hourly agent evaluations for 24 hours before production, and notes classic vs standard tier coverage and false‑flag rates.
AI-written summary · about
chapters (5)
- 0:00Introduction to AI agents
- 3:59Managing safety with guardrails
- 11:15Automated reasoning concepts
- 18:50Policy and formal verification
- 26:13Genesys Cloud Copilot case study
related talks
41mBuild AI applications with Guardrails for Amazon Bedrock
11mBuilding Secure Backend Services with AI Agents
01h 03mBuilding Secure Infrastructure for Productive AI Agents
17mGuardrails First: Designing AI Agents for Security Operations
18mBuild responsible AI applications with Guardrails for Amazon Bedrock
17mWhat is AI Agent Governance?
30mWhy Prompt-Level Guardrails Fail for AI Agents
33mShipping Enterprise AI Agents with Amazon Bedrock AgentCore