Sandbox (Article 57)
Regulatory testing environment for AI
TL;DR
A regulatory sandbox under Art. 57 EU AI Act (Regulation (EU) 2024/1689) is a controlled environment operated by the competent supervisory authority for the development, testing and validation of innovative AI systems.
What is a Sandbox (Article 57)?
Added value of a sandbox:
- Support from the supervisory authority during the development phase
- Building up understanding for SMEs and start-ups
- Reduction of compliance risk before placing on the market
- Reduced fine risks for breaches of obligations
Sandbox status in the DACH region (as at 17 August 2026) — the deadline for setting up the national AI regulatory sandboxes under Art. 57 AI Act has been postponed to 2 August 2027:
- Germany: the national AI regulatory sandbox under Art. 57 AI Act must be operational by 2 August 2027 (deadline postponed from 2 August 2026); an operational status of individual sandboxes is not evidenced in the official sources reviewed
- Austria: the deadline for the national AI regulatory sandbox under Art. 57 AI Act runs until 2 August 2027; Austria had not designated a market surveillance authority under Art. 70 AI Act as at 17 August 2026 — the AI service desk (KI-Servicestelle) at RTR is an information and advisory body without supervisory powers
- Switzerland: no EU AI Act, but a FINMA and Bakom sandbox since 2024
Practical example
A start-up with an AI solution for credit scoring uses the Federal Network Agency (Bundesnetzagentur) sandbox. 12 months of development with support from the Federal Commissioner for Data Protection (BfDI). Output: a completed conformity assessment plus a DPIA and a reduced fine risk when the system is later placed on the market.