Section 1.4 — The LLM Attack Surface

Large language models have introduced a new category of security vulnerability that didn't exist before. Because LLMs process and generate natural language, the line between "instructions" and "content" becomes blurred — and that's exactly what attackers exploit.

In this section we cover prompt injection, jailbreaking, and how to build LLM deployments that are genuinely resilient to manipulation — not just policy-compliant on paper.