Stop harmful, leaked or made-up output before users see it.
Starting with Threat model for LLM apps, Prompt injection, Input filtering, Output checks, Least privilege for tools, Red teaming, and more
In a picture
Stop bad or unsafe requests early.
Limit what the model can do and say.
Validate the answer before anyone sees it.
Stage 1 of 3: Check input
Open the full conceptMade for you now: slide lessons from basics to production, usually with code and a quiz. See a finished one
Stop harmful, leaked or made-up output before users see it.
AI EngineeringStage 13 of 17: Guardrails and security
Start with these 6. 11 more are one tap away.
List what can go wrong: leaks, abuse and wrong actions.
How attackers hijack prompts and tools, and how to defend.
Check and limit what reaches the model.
Grounding, PII and policy checks.
Give the model only the access the task needs.
Try to break your own system before others do.