AI Safety & Guardrails
What guardrails are and how they work, their documented limitations, the attack surface (prompt injection and jailbreaking), red-teaming as an evaluation method, and the layered, defense-in-depth approach to deploying LLMs responsibly.