# Chapter 13 – Where to Go Next > “Learn the rules. Then break them better.” ⚠️ **Disclaimer:** These techniques can get you in trouble. Use responsibly. [Full disclaimer →](../DISCLAIMER.md) --- If you’ve made it this far, you’ve got the skills to: - Build a prompt - Break a prompt - Override instructions - Smuggle payloads past filters - And jailbreak an LLM using only your words But this book was just the beginning. Now it’s time to hit the **real-world labs, challenges, and open-source playgrounds** — where you can level up and see what breaks under fire. --- ## 🧪 Live Prompt Injection Labs ### 🔗 [Gandalf by Lakera](https://gandalf.lakera.ai) - Progressive LLM prompt injection challenge - Each level teaches a new tactic - No signup needed ### 🔗 [Prompt Airlines](https://promptairlines.com/) - Prompt injection challenge in a fictional travel agent - Bypass the AI assistant, cancel flights, change data ### 🔗 [MyLLMBank](https://myllmbank.com/) - Test summarizer and classification vulnerabilities - Prompt injection via form inputs and document uploads ### 🔗 [MyLLMDoctor](https://myllmdoc.com/) - LLM red teaming in a simulated healthcare portal - Try to override guardrails and get the assistant to leak info ### 🔗 [DoubleSpeak](https://doublespeak.chat/#/) - Chat with a model with a hidden persona - Try to jailbreak it and make it speak the truth --- ## 🧰 Open Source Labs & Vulnerable Apps ### 🔗 [AI Goat (by OWASP)](https://github.com/dhammon/ai-goat) - Deliberately vulnerable LLM app for red teaming - Includes multi-task scenarios and role confusion ### 🔗 [Damn Vulnerable LLM Agent](https://github.com/WithSecureLabs/damn-vulnerable-llm-agent) - Multi-agent LLM app with poor prompt design - Great for context switching, recursion, role hijack tests ### 🔗 [Spikee](https://github.com/WithSecureLabs/spikee) - Realistic LLM attack chain simulation platform - Includes phishing, data exfiltration, recon --- ## 🧠 Training Paths & Courses ### 🔗 [Hack The Box – AI Red Teamer Path](https://academy.hackthebox.com/path/preview/ai-red-teamer) - Guided training on LLM threats and defenses - Includes hands-on lab machines and injection exercises ### 🔗 [Prompting Labs by Immersive Labs](https://prompting.ai.immersivelabs.com/) - LLM attack challenges designed for security professionals - Includes evaluation-based tasks and writeups ### 🔗 [Learn Prompting](https://learnprompting.org/docs/prompt_hacking/intro) - One of the best prompt engineering + red teaming resources - Study the *offensive techniques* section for more attack patterns --- ## 🧠 Books, Blogs, and Experts to Follow | Name | Why | |------|-----| | **Donato Capitella** | OG red teamer in LLM security, creator of `ai-goat` | | **Jason Haddix** | Offensive security + LLM threat modeling thinker | | **Kai Greshake** | Research on prompt injection & data leakage | | **Simon Willison** | Developer-focused LLM safety research | | **Embrace the Red** | Blog on red teaming and model behavior risk | | **Sander Schulhoff** | Author of Learn Prompting, community builder | --- ## 👊 What’s Next for You Want to go deeper? - Try writing your own **LLM agent from scratch** - Build a challenge app with **intentional prompt flaws** - Contribute to `ai-goat`, `dv-llm-agent`, or red teaming repos - Start writing **novel jailbreaks** and submit to HackAPrompt - Join Discords, CTFs, or red team slacks and test your stuff You now have the mindset, tools, and tactics to: > **explore prompt space offensively — and defensively.** --- ## 🚀 Thanks for Reading This book was written to get more hackers into AI security — fast, tactical, and fun. If it helped you: - ⭐ Star the GitHub repo - 🔄 Share it with others - 📚 Build something on top of it Red teaming prompts isn’t just fun. It’s necessary. --- > “Words are now code. Prompts are now exploits. > The interface has changed — but the mission hasn’t.” Stay sharp, and keep breaking the box. — Randall