AIG-007 AI System Requirements and Design Documentation
Description
Before development or significant enhancement begins, the intended purpose, success criteria, deployment context, user population, and known constraints of the AI system are documented. Design decisions addressing socio-technical implications (e.g. fairness, explainability, safety modes) are recorded alongside technical specifications. Requirements documentation is version-controlled and approved by the system owner before development proceeds.
Rationale
Undocumented intent makes post-deployment evaluation impossible; design decisions that are not recorded cannot be challenged, revised, or audited.
Applicability (9 profiles)
Obtain the intended purpose, deployment context and design documentation from the provider (AIG-015 documentation, AIG-034 instructions for use). The deployer's own statement of purpose and context for each use is the AIG-039 use register.
Obtain the intended purpose, deployment context and design documentation from the provider (AIG-015 documentation, AIG-034 instructions for use). The deployer's own statement of purpose and context for each use is the AIG-039 use register.
Framework Mappings (13)
| DSP-24 | Data Differentiation and Relevance | informative |
| GRC-13 | Explainability Requirement | informative |
| EU-AI-Art.11.1 | Technical Documentation — Preparation and Maintenance | partial |
| A.6.2.2 | AI system requirements and specification | full |
| A.6.2.3 | Documentation of AI system design and development | full |
| MP-1.1-001 | AI System Purpose and Deployment Context | MP-1.1-001 | full |
| MP-1.1-002 | AI System Purpose and Deployment Context | MP-1.1-002 | full |
| MS-2.12-002 | AI Environmental Impact Assessment | MS-2.12-002 | informative |
| MAP 1.1 | AI System Purpose and Deployment Context | full |
| MAP 1.6 | AI System Requirements Elicitation | full |
| MAP 2.1 | AI System Task and Method Definition | full |
| MAP 3.1 | AI System Benefit Examination | partial |
| MAP 3.3 | AI Application Scope Specification | full |
Evidence (2)
Version-controlled AI system requirements and design documentation created before development commenced, covering intended purpose, success criteria, deployment context, user population, constraints, and socio-technical design decisions.
Example: AI System Design Document · Recommendation Engine v1.2 (GitHub repo docs/design/), approved by Product Owner on 2025-06-01, containing purpose statement, fairness requirements, explainability approach, and known constraints
Test: Request design documentation for a sample of AI systems. Verify: (1) document exists and is version-controlled (Git history or version metadata), (2) intended purpose and success criteria are explicitly stated, (3) socio-technical concerns (fairness, explainability, safety modes) are addressed, (4) the document was approved before development began (compare approval date to first code commit date), (5) known constraints are documented.
Repository and pipeline configuration showing the approved design document as a gate on the branch the system is built from.
Example: Branch protection and pipeline gate export, repo ml-platform/pricing-model, 2026-08-14: required status check design-doc-approved, required reviewer group ai-system-owners
Test: Read the branch protection and pipeline configuration for each repository that builds a production AI system. Verify: (1) a gate exists that blocks a merge to the build branch without an approved design document reference, (2) the approving group resolves to the system owner role rather than to the development team, (3) the gate is enabled on every repository in the AI system inventory rather than on a sample, (4) the configuration records no override of the gate in the period, or each override resolves to a recorded exception with an owner and an expiry date.
Questions (2)
Is a design document approved for each AI system before development begins?
Undocumented intent makes post-deployment evaluation and audit impossible. Look for version-controlled design documents that predate development commencement and are approved by the system owner.
Which of the following are recorded in your AI system design documentation before development proceeds?
All six elements are expected. The socio-technical decisions, fairness, explainability and safety modes, are the ones most often absent from documentation inherited from a general software development template.