Thinking Outside the Technical Standardisation Box: The Role of Standards Under the Draft EU Artificial Intelligence Act

A lack of standardisation is a significant barrier to wider adoption of artificial intelligence (AI) in society. Without the establishment of commonly accepted standards for the development and deployment of AI systems, technical barriers will remain. Unexpected outcomes, in turn, are likely to generate distrust. This will result in significant socio-economic opportunity costs. In response, recent years have seen a proliferation of AI standardisation initiatives across multiple jurisdictions. As part of these efforts, the EU has taken a relatively unique approach by relying on private European standardisation organisations (ESOs) to develop technical standards which are intended to operationalise the legal requirements for certain AI systems contained in its proposed Artificial Intelligence Act. Despite this reliance, this article demonstrates how operationalising these requirements requires answering a series of hard normative questions that ESOs are unlikely to answer. While much scholarly and political attention has focused on reforming and modifying European technical standardisation in anticipation of these shortcomings, this article suggests that we should think outside this ‘technical standardisation box’. Another less explored standardisation process should be considered instead: the development of conventions through regulatory sandboxing.