Overview
ISO/IEC 12792:2025 - Information technology - Artificial intelligence (AI) - Transparency taxonomy of AI systems defines a structured taxonomy of information elements to help stakeholders identify and address transparency needs for AI systems. The document describes the semantics of each element and how these elements map to the transparency objectives of different stakeholders. It is applicable to any organization and any AI application, and its content is organized into context-, system-, model- and dataset-level taxonomies, plus guidance on stakeholder roles and constraints on disclosures.
Key Topics and Technical Scope
ISO/IEC 12792:2025 focuses on standardized information elements rather than prescriptive technical controls. Important topics covered include:
- Stakeholders’ needs & transparency objectives - defining goals and roles relevant to AI transparency (e.g., developers, operators, auditors, consumers).
- Constraints on transparency disclosures - legal, privacy, security or commercial limits that affect what can be disclosed.
- Context-level taxonomy - societal, labour, consumer and environmental context information relevant to system transparency.
- System-level taxonomy - basic system information, governance, risk and quality management, applicability, technical characteristics (inputs/outputs, logging, APIs, deployment, configuration).
- Model-level taxonomy - model purpose, technology type, features, algorithms, training/build processes, hyperparameters, compute requirements and model evolution.
- Dataset-level taxonomy - data provenance, properties, domain details (including language and vision specifics), biases, preparation and maintenance.
- Quality & performance - verification/validation processes, runtime measurements and comparison with alternatives.
- Examples and templates - informative annexes with sample transparency templates and stakeholder role examples.
Practical Applications and Who Would Use It
ISO/IEC 12792:2025 is intended for use by a broad set of AI stakeholders:
- AI developers & data scientists - to document model design, datasets, provenance and model/data limitations.
- Risk managers & compliance officers - to assess transparency-related risks and evidence for audits.
- Product managers & procurement teams - to specify transparency requirements in supplier contracts and evaluate AI vendor claims.
- Auditors & regulators - to interpret and request standardized transparency artefacts.
- Operators & integrators - to implement logging, monitoring and disclosure practices consistent with stakeholder needs.
Use cases include creating consistent model cards, dataset documentation, AI system transparency reports, procurement questionnaires, audit evidence packages, and registries of AI systems.
Related Standards (high level)
ISO/IEC 12792:2025 complements other AI governance and quality efforts by providing a taxonomy for transparency elements that can be integrated with organizational governance, risk management, and AI-specific standards and guidance.
Keywords: ISO/IEC 12792, AI transparency, transparency taxonomy, AI systems, data provenance, model documentation, dataset documentation, AI governance, stakeholder transparency.