<bØt-tri¢ks> | Prompt Injection Compendium
Search
AI Security Learning Center

Learn AI Security by Breaking It

The Prompt Injection Compendium for humans and agents. Lessons, techniques, evasions — and hands-on labs built with Arcanum-Sec. We cover everything AI security, but we thrive on prompt injection.

Featured in Jason Haddix's Attacking AI Course
Bot-Tricks theme artwork
Explore the Compendium
All lessons →
Techniques
Technique

Direct Override

Plain instruction replacement routes and early success patterns.

L1Start here
 Technique

Narrative Injection

Role-play and fictional framing that reshapes the model's prompt reading.

L2Flexible
 Technique

Chunking

Piecewise extraction after refusal when direct requests stop working.

L3Methodical
Evasions
Evasion

Base64

Encoded payload wrappers that change what filters notice first.

E1Encoding
 Evasion

Leetspeak

Light character substitution for beginner-friendly bypass work.

E1Readable
 Evasion

Reverse

Backwards text tricks and lightweight transformations for probing.

E1Lightweight
Fundamentals
Fundamental

First Try Fallacy

Do not stop after one failed prompt -- iterate and observe.

F1Mindset
 Fundamental

Persistence Matters

Iterate deliberately, compare versions, keep observations useful.

F1Process
 Fundamental

How Agents Explore

Stable IDs, markdown access, and retrieval-friendly site structure.

F1Agent-ready
Hands-On Labs × Arcanum-Sec

Practice on Real Scenarios

Train against simulated real-world AI applications. Our partnership with Arcanum-Sec brings you labs inspired by real CVEs — indirect injection through meeting summaries, account takeover chains, and more. New labs shipping regularly.

Arcanum Labs Blind XSS Blind SSRF Heritage Labs
Enter the Labs →

Lesson content shaped by Arcanum-Sec research resources, the Arcanum PI Taxonomy, and our own lab discoveries.