Governance and Responsible AI

Proving the system behaves: model cards and lineage, CloudTrail audit trails, drift and misuse detection, transparency through reasoning displays and agent tracing, fairness evaluation with LLM-as-a-judge, and policy-aligned guardrails.

Content current as of 2026-10.

Lessons

1. Model cards, lineage, and audit trails
2. Transparency: reasoning, confidence, tracing
3. Fairness evaluation and policy alignment

Sources

• AIP-C01 exam guide — Content Domain 3 (tasks 3.3–3.4)
• Amazon SageMaker Model Cards
• Well-Architected Generative AI Lens — enable tracing for agents and RAG
• Amazon Bedrock — model invocation logging