Validating AI-Enabled Systems
Map the application, sources, interfaces and human decisions that shape an AI-assisted outcome.
Read chapterAI Assurance Academy
Learn how to assess and control AI used in quality and regulated processes. Follow practical examples from intended use and risk assessment through testing, human review, and ongoing monitoring.
Browse all chapters
20 of 20 chapters
Part 1 · Chapters 1–5
Define the use and responsibilities
Chapter 1 Chapter 2 AI System Types and Boundaries
Identify the capabilities and permissions that define an AI system’s assessment boundary.
Read chapterChapter 3 Defining Intended Use for AI
Write a bounded intended-use statement with users, inputs, permitted reliance and exclusions.
Read chapterChapter 4 GxP Applicability for AI
Document how actual AI use can affect a regulated activity or record.
Read chapterChapter 5 AI Governance and Inventory
Assign ownership and build an inventory linked to the approved use and current evidence.
Read chapter
Part 2 · Chapters 6–10
Assess risks, data, and suppliers
Chapter 6 AI Risk Assessment
Connect specific AI failures to consequences, controls and a functional risk assessment.
Read chapterChapter 7 AI Validation Evidence Plan
Plan the evidence needed to support each important claim about the proposed use.
Read chapterChapter 8 AI Data Integrity and Lineage
Trace sources, transformations and versions from input to the approved outcome.
Read chapterChapter 9 Reference Data and Expected Results
Prepare independent reference results and protect evaluation data from leakage.
Read chapterChapter 10 AI Supplier Assessment
Assess supplier evidence, service limitations, responsibilities and change arrangements.
Read chapter
Part 3 · Chapters 11–15
Test performance and human review
Chapter 11 Measuring AI Performance
Calculate precision and recall, examine critical misses and interpret the limits of test results.
Read chapterChapter 12 AI Test Strategy
Plan risk-based scripted, unscripted and exploratory tests for representative and challenging cases.
Read chapterChapter 13 Evaluating Retrieval-Augmented Generation
Evaluate retrieval, source selection, citation support and access controls separately.
Read chapterChapter 14 Critical AI Errors and Unsupported Outputs
Detect unsupported statements, omissions and misleading claims of completeness.
Read chapterChapter 15 Human Review and Oversight
Assess whether reviewers detect important AI errors and document evidence supporting human review.
Read chapter
Part 4 · Chapters 16–19
Control records, changes, and ongoing use
Chapter 16 AI Outputs as Controlled Records
Define which AI-related information becomes a controlled record and how it is retained.
Read chapterChapter 17 AI Change Control
Assess changes to models, prompts, sources, interfaces and intended use.
Read chapterChapter 18 AI Monitoring, Incidents, and CAPA
Choose monitoring signals, incident responses and periodic review actions that protect ongoing use.
Read chapterChapter 19 AI Security and Agent Permissions
Test permissions, prohibited actions, interruption and recovery for action-taking AI.
Read chapter
Part 5 · Chapter 20
Work through an assurance decision
Chapter 20 AI Release Decision: Complete Case Study
Work through a connected evidence package and decide whether expanded reliance is supported.
Read chapter
What each chapter includes
A practical explanation, a fictional worked example, an exercise with a suggested answer, expected records or a worksheet, and relevant primary sources. Related videos include captions and transcripts where available.
Educational material; no certification or regulatory approval is conferred. Read the editorial policy.