Evaluate whether AI regulatory sandbox rules account for small-operator resource constraints

Determine whether AI regulatory sandbox participation rules adequately account for the resource constraints of small operators and whether shared infrastructure or cost-sharing should be adopted so that sandbox evidence is not skewed toward well-resourced participants.

Background

The Covenance sandbox pilot found that the inference costs of large-scale testing fell entirely on the participating company. According to the paper, this limited the scale of testing and consequently the generalisability of the resulting evidence.

The authors identify a broader governance question: if small operators cannot afford extensive testing, sandbox evidence may disproportionately reflect the capacities and priorities of well-resourced participants. Possible responses include shared infrastructure or cost-sharing, but the paper leaves both the adequacy of current participation rules and the desirability of these adaptations unresolved.

References

EUSAiR would refer this observation to the AI Office: it raises the question whether sandbox participation rules adequately account for the resource constraints of small operators, and whether adaptations such as shared infrastructure or cost-sharing should be considered so that sandbox evidence is not skewed toward what well-resourced participants can afford to test.

— MARLA: A Conceptual Scaffold for Regulatory Learning under the EU AI Act  (2609.04877 - Buscemi et al., 4 Sep 2026) in Section Case Studies, subsection “Local to National level: a compliance-tech start-up in an AI Regulatory Sandbox pilot,” Learning stage