ISO AI Transparency Taxonomy

Overview

ISO/IEC 12792:2025, titled "Information technology — Artificial intelligence (AI) — Transparency taxonomy of AI systems," is a pivotal international standard developed by the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC) through their joint technical committee, ISO/IEC JTC 1/SC 42. This standard provides a comprehensive and structured taxonomy of information elements specifically designed to assist various AI stakeholders in identifying and addressing their unique transparency needs concerning AI systems. It serves as a foundational document in the evolving landscape of AI governance, aiming to foster greater understanding, trust, and accountability in the design, development, deployment, and operation of artificial intelligence technologies across diverse sectors and applications.

The significance of ISO/IEC 12792:2025 lies in its ability to offer a common language and framework for discussing and implementing transparency in AI. By defining a consistent vocabulary and semantic structure for transparency disclosures, the standard helps bridge communication gaps between technical developers, business leaders, regulatory bodies, and end-users. It acknowledges that transparency is not a monolithic concept but rather a multi-faceted requirement that varies depending on the context, the AI system's purpose, and the specific concerns of different stakeholders. The standard's applicability extends to any organization or application involving an AI system, making it a versatile tool for promoting responsible AI practices globally. Its development underscores the growing international consensus on the importance of transparency as a cornerstone of trustworthy and ethical AI.

Definitions

ISO/IEC 12792:2025 establishes a clear set of definitions crucial for understanding and implementing AI transparency. At its core, the standard defines a "Transparency Taxonomy" as a structured classification system of information elements. This taxonomy is specifically designed to aid AI stakeholders in identifying and fulfilling the necessary transparency requirements for AI systems. It provides a systematic way to categorize and present information, ensuring consistency and comprehensibility across different contexts and applications. The standard delves into the "Semantics of Transparency Elements," which refers to the precise meaning and interpretation of these various information components. Understanding these semantics is vital for ensuring that transparency disclosures are accurate, relevant, and effectively convey the intended information to diverse audiences, thereby avoiding ambiguity and misinterpretation.

Furthermore, the document elaborates on several key taxonomic levels. The "Context-Level Taxonomy" addresses transparency information related to the broader societal, environmental, and organizational context in which an AI system operates. This includes disclosures pertinent to labor impacts, consumer rights, and the overall ethical implications of the AI's deployment. The "System-Level Taxonomy" focuses on transparency details concerning the AI system itself, encompassing basic system information, its governance structures, associated management systems, risk management protocols, and quality assurance measures. It also covers the system's intended applicability, capabilities, limitations, and technical characteristics such such as inputs, outputs, production data, logging, storage, APIs, human factors, deployment, and configuration. Finally, the "Model-Level Taxonomy" provides specific transparency information about the AI model, including its metadata, processing characteristics, dependencies, technology type, extracted features, and details regarding its verification and validation processes. These definitions collectively form a foundational glossary, enabling a standardized approach to AI transparency across the entire AI lifecycle.

Governance and Institutional Framework

The governance and institutional framework surrounding ISO/IEC 12792:2025 is primarily anchored within the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC). Specifically, the standard was developed under the purview of ISO/IEC JTC 1/SC 42, which stands as the world's premier technical committee dedicated to artificial intelligence standardization. This joint technical committee plays a critical role in shaping the global landscape of AI governance by developing comprehensive international standards that promote the safe, ethical, and effective operation of AI systems across various markets. SC 42's mandate is broad, covering the entire AI ecosystem, from foundational concepts and terminology to trustworthiness, risk management, and ethical considerations.

SC 42 operates with an "ecosystem approach," integrating regulatory, business, societal, and ethical perspectives into its standard development process. This collaborative model ensures that standards like ISO/IEC 12792:2025 are not only technically robust but also universally applicable and responsive to the diverse needs of stakeholders worldwide. The committee's work is characterized by consensus-building, involving experts from over 60 countries, fostering international cooperation and alignment in AI governance. By providing guidance to other ISO and IEC committees developing AI applications, SC 42 acts as a central coordinating body, ensuring consistency and coherence across the vast array of AI-related standards. This institutional framework is designed to build trust in AI technologies by providing actionable compliance pathways and promoting responsible AI development and deployment on a global scale.

Key Provisions

The key provisions of ISO/IEC 12792:2025 revolve around its core objective: to establish a structured taxonomy for transparency in AI systems. The standard specifies a comprehensive set of information elements that are critical for achieving transparency, organized at multiple levels to address diverse stakeholder needs and objectives. It moves beyond a simple definition of transparency by providing a detailed framework that categorizes the types of information that should be disclosed and explains their relevance. This includes, but is not limited to, information about the AI system's intended purpose, its capabilities and limitations, the data used for its training and operation, and its technical characteristics. The standard emphasizes that effective transparency requires a clear understanding of the semantics of these information elements, ensuring that disclosures are meaningful and comprehensible to their intended audience.

A significant aspect of the standard's provisions is its multi-layered approach to transparency. It outlines a context-level taxonomy that covers societal, environmental, and organizational implications, including labor and consumer-related disclosures. The system-level taxonomy focuses on the AI system's basic information, governance, management systems, risk and quality management, and its applicability. Furthermore, it delves into technical characteristics such as inputs, outputs, production data, logging, storage, APIs, human factors, deployment, and configuration. The model-level taxonomy provides details on model metadata, processing characteristics, dependencies, technology type, extracted features, and verification/validation processes. These provisions are not prescriptive implementation steps but rather a consistent vocabulary and semantic structure for transparency disclosures. This enables organizations to integrate the taxonomy into their existing management systems and regulatory compliance frameworks, thereby improving consistent and auditable AI transparency.

Scope and Application

The scope of ISO/IEC 12792:2025 is intentionally broad, encompassing any organization or application that involves an AI system. This wide applicability ensures that the standard can be utilized across various industries, sectors, and types of AI deployments, from small-scale applications to large, complex enterprise-level systems. The document's primary function is to define a structured taxonomy of information elements, which serves as a universal tool to assist AI stakeholders in identifying and effectively addressing their specific transparency requirements. This includes, but is not limited to, developers, providers, deployers, users, regulators, and individuals potentially affected by AI systems. The standard recognizes that each of these stakeholder groups may have distinct needs for transparency, and its taxonomy is designed to accommodate this diversity.

In terms of application, ISO/IEC 12792:2025 provides a foundational vocabulary and structural framework rather than a prescriptive set of implementation steps. This approach allows for flexibility in how organizations integrate transparency considerations into their existing AI lifecycle processes and governance structures. It is applicable to both new and existing AI systems, offering guidance on how to systematically categorize and disclose relevant information throughout the AI system's lifecycle. The standard's focus on semantics ensures that the disclosed information is not only present but also clearly understood by all relevant parties. By providing a common reference point for transparency, the standard facilitates better communication, enhances trust, and supports compliance with emerging ethical guidelines and regulatory requirements related to AI. It is designed to be complementary to other international and national AI governance, risk management, and technical standards, offering a neutral and consistent framework that can be integrated into broader AI management systems.

Implementation Framework

The implementation framework for ISO/IEC 12792:2025 is designed to be flexible, allowing organizations to integrate its transparency taxonomy into their existing AI governance and management systems. The standard itself provides a structured vocabulary and semantic framework rather than a rigid set of instructions, enabling organizations to adapt its principles to their specific operational contexts and technological stacks. Organizations are expected to leverage the taxonomy to systematically identify the information elements relevant to their AI systems and the various stakeholders involved. This involves mapping the standard's transparency categories—such as context-level, system-level, and model-level taxonomies—to their internal data collection, documentation, and disclosure processes. The goal is to ensure that transparency is embedded throughout the AI system's lifecycle, from conception and development to deployment and ongoing monitoring.

Successful implementation often involves aligning ISO/IEC 12792:2025 with other complementary AI standards, particularly those from the ISO/IEC 42xxx series, such as ISO/IEC 42001 (AI Management System) and ISO/IEC 42005 (AI System Impact Assessment). For instance, the transparency requirements identified using ISO/IEC 12792:2025 can inform the design and operational controls within an AI Management System established under ISO/IEC 42001. Similarly, impact assessments guided by ISO/IEC 42005 can highlight areas where enhanced transparency is crucial to mitigate risks and ensure ethical considerations are addressed. The standard encourages organizations to establish clear roles and responsibilities for managing transparency, develop internal guidelines for disclosure, and train personnel on the importance and application of the taxonomy. By providing a consistent framework, the standard supports organizations in building auditable and accountable AI systems, thereby fostering greater trust among stakeholders and facilitating compliance with evolving regulatory landscapes.

Monitoring and Evaluation

Monitoring and evaluation within the context of ISO/IEC 12792:2025 are integral to ensuring the ongoing effectiveness and relevance of AI transparency efforts. While the standard itself provides a taxonomy for transparency rather than a prescriptive monitoring framework, its successful application inherently requires organizations to establish mechanisms for continuously assessing how well they are meeting their transparency objectives. This involves regularly reviewing the information elements being disclosed, evaluating their clarity and comprehensibility to target stakeholders, and verifying their accuracy. Organizations should implement feedback loops to gather input from users, affected individuals, and regulatory bodies regarding the sufficiency and utility of the provided transparency information. This continuous feedback is crucial for identifying areas where transparency disclosures might be inadequate, misleading, or simply not meeting stakeholder expectations.

Furthermore, the evaluation process should extend to assessing the internal processes used to generate and maintain transparency information. This includes auditing documentation practices, data provenance, and the mechanisms for updating disclosures as AI systems evolve or their operational context changes. Organizations should consider establishing performance indicators related to transparency, such as the completeness of disclosures, the accessibility of information, and stakeholder satisfaction with the level of transparency provided. Aligning these monitoring and evaluation activities with broader AI management systems, such as those defined in ISO/IEC 42001, can create a cohesive approach to AI governance. Regular reviews by internal and external auditors can help ensure that the application of the transparency taxonomy remains consistent, compliant, and continuously improved, thereby reinforcing trust and accountability in AI systems.

Relationship to Other Instruments

ISO/IEC 12792:2025 does not exist in isolation but is intricately linked to a broader ecosystem of international AI governance instruments, particularly other standards developed by ISO/IEC JTC 1/SC 42. It serves as a complementary standard, providing a foundational taxonomy for transparency that can be integrated with and enhance the effectiveness of other AI-related standards. For instance, it works in conjunction with ISO/IEC 42001:2023, the AI Management System standard, by providing the specific elements of transparency that an organization's AIMS should address. The transparency taxonomy helps operationalize the transparency principles outlined in broader management system standards, ensuring that organizations have a structured approach to fulfilling these requirements.

The standard also has a strong relationship with standards focused on risk management and impact assessment. ISO/IEC 42005:2025, which provides guidance on performing AI system impact assessments, can leverage the transparency taxonomy to identify and disclose information relevant to potential impacts on individuals and society. Similarly, ISO/IEC 23894:2023 on AI risk management benefits from a clear transparency framework, as disclosing risks and mitigation strategies is a key aspect of responsible AI. Furthermore, ISO/IEC 12792:2025 complements technical standards like ISO/IEC TS 6254:2025 on explainability and interpretability, providing the structural elements through which such technical insights can be effectively communicated. It also builds upon foundational documents such as ISO/IEC 22989:2022 for AI concepts and terminology and ISO/IEC 23053:2022 for the AI system lifecycle framework, ensuring a consistent and coherent approach across the entire AI standardization landscape. This interconnectedness ensures that transparency is not an afterthought but an integrated component of a holistic AI governance strategy.

International Alignment

ISO/IEC 12792:2025 plays a crucial role in fostering international alignment in AI governance by providing a globally recognized and harmonized framework for transparency. Developed through a consensus-based process involving experts from numerous countries under the ISO/IEC JTC 1/SC 42 committee, the standard reflects diverse cultural, regulatory, and technological contexts. This collaborative approach ensures that the transparency taxonomy is universally applicable, overcoming potential barriers to cross-border cooperation and the interoperability of AI systems. By establishing a common language and structure for transparency disclosures, the standard facilitates consistent practices across different jurisdictions, which is increasingly vital as AI technologies are deployed globally and subject to varied national and regional regulations.

The standard's emphasis on a neutral vocabulary and semantic structure allows it to be integrated with various national and international AI policy initiatives and regulatory frameworks, such as the EU AI Act or other national AI strategies. It provides a practical tool for organizations operating in multiple regions to demonstrate their commitment to responsible AI, ensuring that their transparency efforts align with global best practices. Furthermore, ISO/IEC JTC 1/SC 42 actively collaborates with other international organizations and standards bodies, ensuring that its work, including ISO/IEC 12792:2025, contributes to a cohesive global approach to AI governance. This alignment is essential for building public trust in AI, promoting innovation while mitigating risks, and supporting the ethical deployment of AI technologies worldwide, ultimately contributing to broader societal goals and sustainable development.

Implementation Timeline

Adoption and Endorsement

Sources and References