AI & LLM Security

Learn to find and fix the vulnerability classes unique to LLM-integrated systems: direct and indirect prompt injection, insecure output handling that reintroduces classic injection one layer removed, training-data memorization and leakage, and building guardrail pipelines whose effectiveness is measured with a red-team suite instead of assumed.

Builds on: Python 3.12 (regex-based injection detection, SQLite-backed insecure-output demos, canary-based memorization checking, guardrail pipeline + red-team scoring)