Security is the sixth step of the design method, and it is a design step: the controls that matter most in an AI system are architectural — which component can see what, and which can do what — and they cannot be added after the boxes are drawn.
| # | Lesson | The question it answers |
|---|---|---|
| 01 | Trust Boundaries in an AI System | Where does untrusted text enter, where do effects leave, and what lies between? |
| 02 | A Secure Reference Architecture | What does a design look like with every control in place? |
These two lessons give a designer what is needed at the whiteboard. The full treatment — how each attack works, how to secure each stage of the lifecycle, and how to secure agents — is its own course: AI Security Engineering.