Menu
Course
AI Security & Guardrails
Threat-model LLM apps — prompt injection, RAG poisoning, agent scoping, harness authorization, and adversarial evals in CI.
0/3 modules complete0%
Modules
3 total
Injection And Trust Boundaries
Explain why LLMs cannot reliably separate instructions from data — the root cause of prompt injection.
Continue
2RAG, Vectors, And Data Exposure
Treat every retrieved chunk as untrusted input — not ground truth.
Open
3Agents, Controls, And Security Evals
Scope tools with least privilege — load only what the current task needs.
Open