Prompt injection and LLM security — companion videos

LiveOverflow

Frames prompt injection as a classic injection attack against systems that mix instructions and untrusted data — with a concrete content-moderation example where an attacker frames an innocent user. The mental shift from "the model is the target" to "the application is the target" is exactly the move the article opens with.

What you should get from this: Model prompt injection as untrusted-data mixing and design boundaries around tool use.

Watch or know first: General web-security intuition helps; no prior LLM-security knowledge assumed.

LiveOverflow

Walks through the actual defence-in-depth playbook — taint analysis on LLM output, restricting expected output shapes, user isolation, few-shot scaffolds, fine-tuning, temperature 0 for determinism, redundancy for critical paths. It matches the article's defence-stack section almost item for item.

What you should get from this: Review prompt-injection defenses such as taint analysis, output-shape restrictions, user isolation, deterministic settings and redundant checks for critical paths.

Watch or know first: Watch the attack-side video first — the defence playbook assumes that threat model.