AIG-015 AI System Technical Documentation
Description
Technical documentation is created before an AI system is deployed and kept current throughout its lifecycle. Documentation covers: system architecture and components, intended purpose and use cases, performance metrics and limitations, training data summary, known failure modes and edge cases, human oversight mechanisms, hardware and compute requirements, and update/maintenance schedule. Documentation is made available to deployers, operators, and competent authorities on request. For GPAI models, a public training data summary is published per regulatory requirements.
Rationale
Technical documentation is the primary evidence artefact for AI governance audits; undocumented systems cannot be assessed for compliance.
Framework Mappings (6)
| EU-AI-Art.11.1 | Technical Documentation — Preparation and Maintenance | full |
| EU-AI-Art.18.1 | Documentation Retention — 10-Year Retention Obligation | partial |
| EU-AI-Art.53.1 | GPAI Model Obligations — Technical Documentation | full |
| EU-AI-Art.53.4 | GPAI Model Obligations — Training Data Summary | full |
| A.6.2.7 | AI system technical documentation | full |
| MEASURE 2.8 | AI Transparency and Accountability Risks | partial |
Evidence (1)
AI system technical documentation for each production system, covering system architecture, intended purpose, performance metrics, known limitations, training data summary, failure modes, and human oversight mechanisms.
Example: Technical Documentation — AI Underwriting Assistant v3 (Confluence, version-controlled), covering architecture diagram, intended use cases, accuracy/recall metrics, 8 documented known limitations, training data summary, fail-safe behaviour description, and instructions for oversight personnel
Test: Request technical documentation for a sample of production AI systems. Verify: (1) all required sections are present (architecture, purpose, metrics, limitations, training data summary, failure modes, oversight mechanisms), (2) documentation is version-controlled and the current version matches the deployed model version in the registry, (3) documentation is retrievable on request (not locked in individual engineer knowledge), (4) for GPAI models, confirm a public training data summary exists.
Questions (2)
Is technical documentation created before deployment and kept current for each AI system throughout its lifecycle?
Technical documentation is the primary evidence artefact for AI governance audits and regulatory inspections. It must be version-controlled and the current version should correspond to the deployed model version.
Which of the following sections are included in your AI system technical documentation?
All seven sections should be present. 'Known failure modes' and 'human oversight mechanisms' are most commonly absent from inherited software documentation templates and are the sections most scrutinised by enterprise buyers.