Specify layered, testable and costed guardrails
A single guardrail placed in front of the model does not protect an AI feature. This lesson teaches you to spread protections across several levels, separate the ones that truly cap a risk from those that only make it less likely, and specify each one with a test and a cost.
Lesson objective
By the end of this lesson, you will be able to choose, for each risk in the register, guardrails in several layers (input, context, actions, output, operations), to tell those that bound the risk from those that reduce it, and to specify them with a testable criterion and a cost.
Topics covered
- AI guardrails
- defense in depth
- guardrails
- security requirements
In the glossary
Where it fits
Guardrails and security operations
Which guardrails do you specify, how do you know they hold, and what do you do on incident day?
Lessons in this module
- Specify layered, testable and costed guardrails (this lesson)
- Red teaming, abuse monitoring and incident response
What you will learn in this module
This lesson is part of the course Secure an AI product: prompt injection, guardrails and the AI Act
- Specify layered guardrails (input, output, rights, human approval, isolation, limits) with testable criteria and their cost.
- Plan red teaming, abuse monitoring and incident response for an AI feature.
Related courses
- AI & ProductAdvancedRegular practice: you already use the tools or work on this topic.
Build an AI assistant for your product
3 hr
- AI & ProductAdvancedRegular practice: you already use the tools or work on this topic.
Design a RAG architecture that fits your product
3 hr 30 min
- AI & ProductAdvancedRegular practice: you already use the tools or work on this topic.
Evaluate an AI feature: test sets, metrics and LLM judges
3 hr