The European organisations CEN (European Committee for Standardisation) and CENELEC (European Committee for Electrotechnical Standardisation), as well as ETSI (European Telecommunications Standards Institute), are developing harmonised standards that support the AI Act, and which include best practices that may be of interest to organisations wishing to govern, manage and ensure the quality of their Artificial Intelligence systems.

On CEN/CENELEC's website, you can find all the standards defined by the Joint Technical Committee JTC21, which is responsible for standardising Artificial Intelligence. Some of these consist of the direct adoption of existing ISO/IEC standards, whilst in other cases new standards are being developed specifically designed to support the AI Act.

As with ISO/IEC, standards covering basic concepts can be found, such as EN ISO/IEC 22989 Information technology — Artificial intelligence — Artificial intelligence concepts and terminology, EN ISO/IEC 23053 Framework for Artificial Intelligence (AI) Systems Using Machine Learning (ML), or the Draft European Standard (prEN) for characterising the methods and capabilities of artificial intelligence systems (prEN ISO/IEC 42102 Information technology – Artificial intelligence – Framework for characterising AI system methods and capabilities).

1. Governance of AI Systems

As regards governance-related aspects, CEN/CELENEC have not adopted the ISO/IEC 38507 standard, but they have adopted another standard relating to the risk associated with AI systems, such as EN ISO/IEC 23894 Information technology — Artificial intelligence — Guidance on risk management, which is supplemented by prEN 18228 AI Risk Management, which sets out requirements and provides guidance for the risk management of AI systems. In addition, they are developing a checklist (AI Risks – Checklist for AI Risk Management) and a guide to risk management in critical digital infrastructure (Guidance on the Application of Risk Management in Critical Digital Infrastructure).

Another area that CEN/CENELEC explores in depth is ethics, adopting standards such as EN 18274 Competence requirements for professional AI ethicists, and developing others such as Guidance for upskilling organisations on AI ethics and social concerns and the Guidelines on tools for handling ethical issues in the AI system life cycle.

As is to be expected with regulations relating to the AI Act, the legal aspects are set out in standards such as prCEN/CLC/TR 18347 Overview and architecture of standards in support of the EU AI Act and a standard on impact assessment is even currently being developed (Impact assessment in the context of the EU Fundamental Rights).

2. Management of AI systems

In addition to adopting some of the ISO/IEC standards on AI Management Systems (AIMS), such as EN ISO/IEC 42001 Information technology — Artificial intelligence — Management system or the prEN ISO/IEC 42006 Information technology — Artificial intelligence — Requirements for bodies providing audit and certification of artificial intelligence management systems, CEN/CENELEC have developed the standard EN 18286 Artificial intelligence – Quality management system for EU AI Act regulatory purposes, with the aim of helping organisations to comply with the regulatory requirements applicable primarily to those that place high-risk AI systems on the market or put them into service. This EN 18286 standard and EN ISO/IEC 42001 have overlaps and differences which we have already discussed in a previous article.

3. Quality of AI systems

3.1. Data quality of AI systems

Starting with data quality, CEN/CENELEC have adopted both the ISO/IEC standard on the data life cycle (EN ISO/IEC 8183 Information technology — Artificial intelligence — Data life cycle framework), as well as the ISO/IEC 5259 series of standards:

• EN ISO/IEC 5259-1 Artificial intelligence — Data quality for analytics and machine learning (ML) Part 1: Overview, terminology, and examples
• EN ISO/IEC 5259-2 Artificial intelligence — Data quality for analytics and machine learning (ML) Part 2: Data quality measures
• EN ISO/IEC 5259-3 Artificial intelligence — Data quality for analytics and machine learning (ML) Part 3: Data quality management requirements and guidelines
• EN ISO/IEC 5259-4 Artificial intelligence — Data quality for analytics and machine learning (ML) Part 4: Data quality process framework

As we know, the 5259 family of standards helps to define, measure, manage and govern the quality of data used in analytics and machine learning, ensuring that the results are reliable, comparable and auditable. It is also essential for both AI Act compliance and for AI management and quality systems.

In addition, they have published the technical report CEN/CLC/TR 18115 Data governance and quality for AI within the European context, which provides an overview of standards relating to artificial intelligence, focusing on data and its life cycles; and they are developing the standard prEN 18284 Artificial Intelligence — Quality and governance of datasets in AI, which provides guidance and requirements for the creation and management of datasets in the field of artificial intelligence, including design decisions, data collection and data preparation.

3.2 Software quality of AI systems

CEN/CENELEC has adopted the ISO/IEC 25059 standard as EN ISO/IEC 25059 Software engineering — Systems and software Quality Requirements and Evaluation (SQuaRE) — Quality model for AI systems, which, together with software life-cycle processes, are fundamental to ensuring, evaluating and certifying the quality of AI software.

3.3 Model quality of AI systems

The quality of the AI models themselves has also been standardised by CEN/CENELEC, for example:

• For fairness and bias, it has adopted the ISO/IEC standards CEN/CLC ISO/IEC/TR 24027 Information technology – Artificial intelligence (AI) – Bias in AI systems and AI-aided decision-making and CEN/CLC ISO/IEC/TS 12791 Information technology – Artificial intelligence – Treatment of unwanted bias in classification and regression machine learning tasks; and has proposed prEN 18283 Artificial Intelligence – Concepts, measures and requirements for managing bias in AI systems, which defines concepts, indicators and requirements for the assessment and management of bias.
• For transparency, it has adopted EN ISO/IEC 12792 Information technology – Artificial intelligence (AI) – Transparency taxonomy of AI systems
• For Reliability and Robustness, it has adopted CEN/CLC ISO/IEC/TR 24029-1 Artificial Intelligence (AI) – Assessment of the robustness of neural networks – Part 1: Overview, and is developing specific standards such as prEN 18229-1 AI trustworthiness framework – Part 1: Logging, prEN 18229-2 AI trustworthiness framework – Part 2: Transparency, and prEN 18229-3 AI trustworthiness framework – Part 3: Human oversight.

3.4 Quality of AI systems as a whole

As regards other quality characteristics that affect the AI system as a whole, CEN/CENELEC are working on the following:

• Cybersecurity, for which they define the standard prEN 18282 Artificial intelligence – Cybersecurity specifications for AI Systems, which addresses organisational and technical solutions designed to ensure the cybersecurity of high-risk AI systems throughout their entire lifecycle.
• Functional Safety, for which they adopt the ISO/IEC technical specifications of the 22440 family: prCEN/CLC ISO/IEC TS 22440-1 Artificial intelligence — Functional safety and AI systems Part 1: Requirements, prCEN/CLC ISO/IEC TS 22440-2 Artificial intelligence — Functional safety and AI systems Part 2: Guidance, and prCEN/CLC ISO/IEC/TS 22440-3 Artificial intelligence — Functional safety and AI systems Part 3: Examples of application.
• Sustainability, which is addressed in CEN/CLC/TR 18145 Environmentally Sustainable Artificial Intelligence, which establishes a framework for quantifying the environmental impact of AI and its long-term sustainability; and prEN 18287 Artificial Intelligence – Requirements and guidance for the environmental impact evaluation of artificial intelligence systems and services, which sets out the principles and framework for measuring the environmental impact of artificial intelligence systems and services, and provides guidance on reducing that impact throughout their entire life cycle.

4. Other standards

4.1 Conformity standards

In addition to the above standards on assessment and accreditation, CEN/CENELEC have developed two specific standards on conformity: CEN/CLC/TR 17894 Artificial Intelligence – Artificial Intelligence Conformity Assessment, which reviews the methods and practices for assessing the conformity of products, services, processes, management systems, organisations or individuals, insofar as they are relevant to the development and use of AI systems, and prEN 18285 AI Conformity assessment framework, which establishes a correspondence between conformity assessment activities and the different phases of the AI systems life cycle.

4.2 Standards for specific areas

In addition, specific standards are defined for natural language processing (prEN ISO/IEC TR 23281 Artificial intelligence — Overview of AI tasks and functionalities related to natural language processing, and prEN ISO/IEC 23282 Artificial Intelligence — Evaluation methods for accurate natural language processing systems) and computer vision (prEN 18281 Artificial Intelligence – Evaluation methods for accurate computer vision systems, and prEN 18288 Artificial Intelligence – Taxonomy of AI tasks in computer vision).