Skip to main content

ISO/IEC 5259

A family of international standards that sets out how to define, measure, manage, and govern data quality for AI systems, from shared vocabulary to process and governance models.

The Simple Version

A family of international standards that sets out how to define, measure, manage, and govern data quality for AI systems, from shared vocabulary to process and governance models.

Detailed Explanation

ISO/IEC 5259 comprises five published or in-development parts, each addressing a distinct layer of data quality for AI: Part 1 (terminology and overview), Part 2 (data quality measures), Part 3 (data quality management requirements), Part 4 (data quality process framework), and Part 5 (data quality governance framework). The series provides the standards infrastructure needed to satisfy EU AI Act data governance requirements for high-risk systems, offering a coherent framework from vocabulary through operational processes to organisational governance. Parts align with and extend ISO/IEC 25000 (SQuaRE) quality concepts for AI-specific needs.

Key Characteristics

  • Five-part series covering vocabulary, measures, management, process, and governance
  • Developed by ISO/IEC JTC 1/SC 42, the AI standards committee
  • Provides the standards basis for EU AI Act data governance compliance
  • Aligns with and extends ISO/IEC 25000 SQuaRE quality concepts for AI

Why It Matters

AI data governance programmes can use ISO/IEC 5259 as a comprehensive reference architecture, mapping each part to different governance layers from technical quality measurement to executive accountability frameworks.

Real-World Analogy

Like the ISO 9000 series for quality management, a coherent family of standards addressing vocabulary, requirements, guidance, and specific techniques at different levels of detail.

Common Misconceptions

  • ISO/IEC 5259 duplicates ISO/IEC 25012 data quality concepts. 5259 extends and contextualises existing quality standards specifically for AI and analytics use cases.
  • Implementing one part of the series satisfies the whole, each part addresses a different layer; a complete programme typically needs all five parts.

Related Terms

Related Articles

Sources & Further Reading