IFOW Sandbox - Phase 1 Methodology Report

In the last few years, sandboxing has moved from being a peripheral, barely mentioned concept, to a mainstay of regulatory strategy. Yet this solution to AI governance challenges is deployed across highly divergent regulatory and political environments, from jurisdictions where prescriptive AI law has been introduced, to AI superpowers using sandboxing to drive market-led deregulation, through to contexts such as we have in the UK, where there is more interest in understanding the role of existing law and taking a developmental approach.

This diverse set of applications begs the question: what is a sandbox, and what should a sandbox do? We seek to address this question in this paper via lessons we have learned from the work of the IFOW Sandbox - the first (to our knowledge) established to explore the intersection of algorithmic technologies and the multi-regulator domain of the workplace. We articulate the case for this Sandbox, and its guiding design and methodological principles, following its first year of operation.

We also present the rationale for a new approach to sandboxing, and then share our own process, considering the development of our Open Calls, admission criteria, recruitment of participants, and terms of participation. Finally, we discuss our methodology, both in terms of our approach to data analysis and the case study approach taken to engagement with firms.

We are grateful to Trust for London for funding this work.