Learn prompt injection basics, output filters, and practical guardrails for AI products.
Includes practical lessons, exercises, and a clear next step after you enroll.
About this course
Curriculum
Lessons are listed in order. Each one has a goal, a walkthrough, practice, and a short check before you continue.
- 01The threat model for an LLM featurePrompt injection, data leak, and tool abuse.
- 02Trust boundaries and untrusted textUser content is data, not instructions — delimit it.
- 03Output filters and allowlistsSchema, banned destinations, and length.
- 04Secrets, logs, and trainingKeys out of prompts; logs redacted; training off if required.
- 05Abuse tests you run before launchJailbreaks, exfil tricks, and tool-spam.
- 06Incident response for model featuresWho to call when it says something awful or leaks.
- 07Workshop: a complete pass on a real taskCombine the earlier lessons into one working example you can keep.
- 08Quality bar, common mistakes, and a 30-day planLock in a checklist so the skill survives after the course ends.
- 09Measure results so you know the work is actually betterAdd a simple evaluation so quality and cost are not guesswork.
- 10Ship it: handoff, monitoring, and the next projectPut the workflow where other people can use it without you in the room.