CourseLarge Language Models: The Complete Course · Module 11: Safety, Security, and Alignment · part 79 of 103

Part 79 · Module 11: Safety, Security, and Alignment

Lesson 3: Defences against injection, each one measured

23 min read·13 Oct 2026

Module 11 · Lesson 3 · About 55 minutes · Needs: Module 11 Lesson 2 (the attack surface, T-1034); Module 6 Lesson 3 (scopes, the guard chain) helps

The situation

It is Thursday 23 July 2026, week 8. In Monday's red-team session, T-1034 from Leo Park asked the assistant to refund 500 USD "now" and "not ask a human". The role check in code refused it: Leo is a member, not an owner.

T-1034 is a short refund request followed by an instruction aimed at the assistant (red means attack). In week 8 one check in code (green, safe) refused it.

Anika asks: "One check held. What if it had a bug?" You will add five defences, measure each on T-1034 and the 36 shared tickets, and count what each costs in genuine customers wrongly flagged. Two disguised copies of T-1034, simulated in this lesson from the same words, test them.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.

Buy this course · ₹399 See what is inside