Cipher CadetBeginner10 daysFree
LLM Prompt Injection and Jailbreaks: A 10-Day Beginner Course
In ten days you go from never having attacked a chatbot to testing and hardening an LLM application with a method you can repeat.
Taught by Mr Elite 1 of 10 days live A new day every day
I built this course for people who have never done security work. Each day is one lesson of about 45 to 75 minutes, including a hands-on mission. We start with how a language model reads its instructions, then move through direct injection, indirect injection, jailbreaks, system prompt extraction and attacks hidden in images. We finish with defences and a testing method.
Every practical step runs against your own notes or a named SecurityElites practice lab. You never point anything at a real third-party system. I explain each new term the first time it appears, I tell you what we are doing before we do it, and I tell you what just happened afterwards. Each day builds on the one before, and every module ends with a day where you apply what you learned.
By the end you can
- Explain why a language model cannot reliably tell instructions from data.
- Perform and recognise direct prompt injection against a practice chatbot.
- Perform and recognise indirect injection through documents and tool results.
- Describe the main jailbreak families and explain why each works and where each fails.
- Extract a system prompt from a practice lab and rewrite one so a leak costs nothing.
- Explain how text hidden in images and invisible characters carries an injection.
- Run a repeatable test plan against an LLM feature and write up findings with fixes.
Who this is for
- Complete beginners in security who can use a computer and a browser.
- Developers and testers who are starting to build on or around chatbots and want to know how they get attacked.
- Curious students who want a safe, structured first step into AI security.
- Anyone who needs to judge whether an AI feature at work is safe enough to ship.
Before you start
- Nothing beyond being able to use a computer and a web browser.
- A text editor or notes app to record your findings.
- Willingness to write short prompts and read the replies carefully.
- A modern web browser
- A free SecurityElites account for the browser labs
- A plain text editor or notes app
- A screenshot tool (built into your operating system)
Syllabus
One day at a time. Finished days get a tick. Days that are not out yet show when they unlock.
Module 1 Foundations and Direct Injection
Understand how LLM applications are built and break one with plain direct prompt injection.
- 1Orientation: How an LLM App Reads InstructionsBy the end of today you can describe the parts of an LLM application and make a practice chatbot ignore its brief once.Lab: AI Prompt Injection
- 2Direct Prompt Injection: Overriding the BriefBy the end of today you can write and compare several direct injection prompts and say why one worked and another failed.Practice | Lab: AI Instruction Conflict | Coming soon
- 3Role Confusion and Refusal SuppressionBy the end of today you can explain and demonstrate how fake role markers and forced reply prefixes steer a model.Practice | Lab: AI Role-Confusion Injection | Lab: AI Refusal Suppression | Coming soon
Module 2 Indirect Injection and Hidden Channels
Show how hostile content in documents, images and hidden characters controls a model that never saw your prompt.
- 4Indirect Injection: Hostile Content in DocumentsBy the end of today you can plant an instruction inside a document and show a summariser obeying it.Practice | Lab: AI Indirect Prompt Injection | Lab: What the Assistant Reads | Coming soon
- 5Poisoned Knowledge and Tool ResultsBy the end of today you can explain how a poisoned knowledge base or tool output hijacks an assistant, and demonstrate one case.Practice | Lab: AI RAG Poisoning | Lab: AI Agent Hijacking | Coming soon
- 6Hidden Channels: Images and Invisible CharactersBy the end of today you can show how text hidden in an image or in invisible characters carries an injection, and say how to spot it.Review | Lab: AI Visual Prompt Injection | Lab: AI Hidden Character Injection | Coming soon
Module 3 Jailbreaks and System Prompt Extraction
Learn the main jailbreak techniques, their limits, and how system prompts leak.
- 7Jailbreak Techniques and Their LimitsBy the end of today you can name the main jailbreak families, demonstrate two against a practice lab, and explain why each is unreliable.Practice | Lab: AI Jailbreak | Lab: AI Jailbreak | Coming soon
- 8System Prompt Extraction and Prompt HardeningBy the end of today you can extract a system prompt from a practice lab and rewrite one so that leaking it does little harm.Review | Lab: AI System Prompt Extraction (Guided) | Lab: Nothing Worth Leaking | Coming soon
Module 4 Defence, Testing and Capstone
Apply layered defences and a repeatable test method to a whole application.
- 9Defences and a Testing MethodBy the end of today you can choose layered defences for an LLM feature and write a repeatable test plan for it.Practice | Lab: Whose Instruction Was That? | Lab: Who Approved That? | Coming soon
- 10Capstone: Assess an LLM ApplicationBy the end of today you can run a full assessment of a practice LLM application and report findings with fixes.Capstone | Lab: AI Agent Hijacking | Lab: AI System Prompt Extraction | Coming soon
The final project
You take a practice LLM application through a full assessment: you test it for direct, indirect and extraction weaknesses, record each result in a findings table, and propose a fix for each. You finish with a short report that another person could use to repeat your tests.
Questions people ask
Do I need to know how to code?
No. Every mission uses a browser and a notes file. Where a command or code snippet appears, I explain what it does and show the expected result.
Is it legal to try these attacks?
Only against systems you own or have permission to test. In this course that means your own notes and the SecurityElites practice labs, and nothing else. Never aim these techniques at a real service.
Will the jailbreaks from the course still work next month?
Some will not, and that is part of the lesson. I teach why each technique works and why it fails, so you can reason about new variants instead of memorising prompts.
What can I do after the ten days?
You can test a simple chatbot feature, write up findings and suggest sensible fixes. Day 10 ends with pointers to deeper topics such as agents, tool servers and automated red-team tools.