Prompt Injection & Jailbreak Defence
Learn how people trick AI systems into ignoring their instructions, by doing it yourself in a safe practice app. You will run 40 practice attacks against a help assistant, through the chat and through hidden instructions in documents, watch what leaks, then add five defences built into the app and measure how many attacks still get through. You finish with a repeatable attack test that runs before every change. Defensive and educational: the attacks are described, not written out, and only run against the course's own practice app. All practice data is made up.
"We know how our assistant can be tricked, we have defences in layers, and we can show how many known attacks still get through"
6 Interactive Sessions
Short, interactive sessions: watch it work, try it, then build it yourself. You can go deeper at any time in the written chapters.
- 1
How an AI reads instructions
An AI assistant reads its instructions, the customer, documents and tool results as one long text. It cannot reliably tell who wrote each line, and that is where prompt injection starts.
- 2
Direct attacks and jailbreaks
Run the 40 practice attacks against the unprotected app and read what happened. With no defences, 38 get through, and the two that fail are stopped by a rule in code.
- 3
Hidden instructions in documents
An attacker does not need to talk to the assistant. A line planted in a document, an uploaded file or a claim note reaches the model when an innocent customer asks a normal question.
- 4
Defences that help
Five defences built into the app, switched on one at a time, take the practice attacks from 38 getting through to 7. Each one helps. None of them, alone or together, gets to zero.
- 5
When the AI can act
Once an assistant can use tools, a fooled model does more than say the wrong thing. Give each tool a rule: what it may do alone, what needs a person, and what the code checks every time.
- 6
Keep testing
Defences slip when the app changes. Run the attack test before every change, compare the count with the last run, and block a change that lets more through.
Skills you'll practise
Practice data included
- Ask Corri, a practice help assistant with a planted secret in its instructions
- 12 policy documents, 2 carrying hidden instructions
- 40 attack records in 6 kinds
- 3 made-up customers and their claims
What you walk away with
Shareable portfolio
A public URL showing your module timeline, patterns mastered, and completion status.
Everything you build
Download all your work as a ZIP file. It is yours to keep.
Module walkthrough
Each module documented with deliverables and the production pattern you implemented.
Ready to start Prompt Injection & Jailbreak Defence?
First course free. $20 per course after that.