01·Defense-in-depth architecture: pre-LLM filter — model — post-LLM filter — monitoring

Introduction

Effectively securing an LLM-based system requires multiple independent layers of defense — no single control is sufficient. This lesson covers the defense-in-depth architecture: from input filtering (pre-LLM filter), through built-in model mechanisms, to output sanitization (post-LLM filter) and continuous monitoring, focusing on each layer's role, its limitations, and common implementation mistakes.