Imagine an AI system capable of analysing medical images to provide early diagnoses, or an AI embedded in a self-driving car. Behind their operation lies a crucial question: how can we ensure their safety, transparency, and compliance with European law? The answer lies in two words: harmonised standards.<\/p>\n\n\n\n
With the upcoming enforcement of the AI Act<\/a> for high risk products, the European Union sets out clear requirements to protect its citizens while fostering innovation. For companies, applying harmonised AI standards is the simplest way to demonstrate compliance.<\/p>\n\n\n\n The AI Act requires that all high-risk AI systems meet a set of essential requirements. These requirements, developed by industry and market stakeholders, are technically translated into harmonised standards, published in the Official Journal of the EU. Once applied, these standards will confer a presumption of conformity on the manufacturer, meaning that his products will be presumed to comply with the AI Act unless there is evidence of nonconformity compliant\u2014unless evidence to the contrary is found\u2014greatly facilitating the CE marking process of AI systems.<\/p>\n\n\n\n Harmonised standards will provide companies with essential guidelines on how to assess and mitigate risks related to AI products and will offer precise technical specifications to clarify regulatory requirements<\/p>\n\n\n\n ISO standards<\/a> for AI, already internationally recognised, cover many dimensions: risk management, transparency, data governance<\/a>, etc. EU standardisation bodies have reviewed existing standards that could help meet AI Act requirements and have used them as a basis to develop new harmonised standards. Still, in Europe, they are not always sufficient.<\/p>\n\n\n\n Companies using AI tools that fall within the scope of the AI Act must review the existing standards they can apply. They must also follow the work programme and the development of new harmonised standards in Europe that may affect their products.<\/p>\n\n\n\n A key example? ISO\/IEC 42001, which covers AI management systems. It\u2019s valuable for organisational governance but is not directly eligible as a harmonised standard:<\/p>\n\n\n\n The challenge is to better align<\/strong> the work of CEN\/CENELEC<\/strong> with ISO<\/strong> to avoid redundancies. ISO standards should be leveraged but adapted<\/strong> to meet the EU AI Act’s specific requirements.<\/p>\n\n\n\n That said, companies already complying with international standards\u2014such as ISO\/IEC 42001\u2014will be better positioned<\/strong> to adapt to the AI Act and benefit from their early efforts.<\/p>\n\n\n\n As of now, seven European standards<\/strong> are being drafted to support implementation of the AI Act.<\/p>\n\n\n\nHarmonised AI Standards: A Guarantee of Compliance with the EU Framework<\/strong><\/h2>\n\n\n\n
Concrete Benefits:<\/strong><\/h3>\n\n\n\n
\n
ISO Standards and European Norms: Towards Strategic Alignment<\/strong><\/h2>\n\n\n\n
\n
The 7 Harmonised AI Standards in Progress: What Do They Cover?<\/strong><\/h2>\n\n\n\n