ISO 42001 Model Development

Overview

Draft model development documentation for ISO/IEC 42001:2023 (architecture, training, testing, limitations).

Getting Started

Add your source materials to the project workspace, then activate the ISO 42001 Model Development skill. The agent will guide you through the process step by step.

Example conversation

You: We need to document our model development for the recommendation engine—architecture, training data, testing, limitations, and security—for the auditor.

Agent: I’ll draft the model development documentation using the skill reference. I’ll describe architecture and design decisions, training data sources and preparation, testing results (functional, bias, robustness), limitations and known risks, transparency/explainability methods, and information security safeguards. I’ll run model_development_check on the document and fix any missing elements, then link to your data governance and risk register.

Output excerpt

Model development documentation excerpt

Architecture — [e.g. two-tower retrieval + ranking model]. Design decisions: [e.g. offline training, online inference]; [e.g. feature set and constraints]. Assumptions: [e.g. stationarity of behaviour within retrain window].

Training data — Sources: [e.g. engagement events, segment attributes]; period [X]. Preparation: [e.g. filtering, aggregation]; see data governance for provenance and quality. No special-category data.

Testing — Functional: [e.g. accuracy, coverage]. Bias: [e.g. demographic parity, segment fairness]. Robustness: [e.g. stress tests]. Results summarized in [e.g. model card or test report]; limitations documented below.

Limitations — Cold-start; possible bias in long-tail segments; not validated for [e.g. eligibility or safety-critical] use. Mitigations: monitoring, retrain cycle, human oversight.

Transparency — [e.g. Model card]; in-app disclosure; [e.g. feature contribution or surrogate explainability] where requested.

Security — [e.g. model and data access control; pipeline in secure env; no model export to untrusted clients.]

Extension and validation

The skill includes model_development_check, which validates model development documentation for required elements: model architecture and design decisions; training data sources and provenance; testing results (functional, bias, robustness); limitations and known risks; transparency/explainability methods; information security safeguards. Run it after drafting and address any missing elements.