AI regulatory sandboxes: can I test my AI safely under supervision?
Adopted 2026-06-16 · ≈ 2 min read · 
edited by Dirk Baaijen
Yes. By 2 August 2026 at the latest every Member State must offer at least one AI regulatory sandbox: a controlled environment to develop and test innovative AI under supervision, with guidance. It does not exempt you from liability.
Short answer: Yes, that is exactly what AI regulatory sandboxes are for. The AI Act requires every Member State to set up at least one sandbox by 2 August 2026 at the latest: a controlled environment in which you can develop, train, validate and test innovative AI under the supervision of the competent authority. You get guidance, but you remain liable yourself.
What a sandbox is
An AI regulatory sandbox (Art. 57) is a controlled environment set up by the supervisory authority that facilitates the development and testing of innovative AI systems for a limited time, following an agreed plan. The idea is to establish, together with the authority, how your system relates to the rules before you go to market.
- Guidance: you work under supervision and receive advice on complying with
the AI Act and related EU rules.
- Legal certainty: the authority provides written feedback, for example on how
the rules apply to your system. This gives you a foothold, but not a blanket approval.
- Duration and exit: participation is temporary and ends according to the
plan; afterwards you complete the steps towards market entry yourself.
What a sandbox does not provide
- No exemption from liability. If your AI causes harm during the sandbox, you
remain responsible for it. The sandbox does not suspend applicable law.
- No automatic conformity. Participation is not a seal of approval; you still
have to meet all requirements yourself before going to market.
Support for SMEs and start-ups
The AI Act provides extra support for smaller players (Art. 62): priority access to the sandbox if you meet the conditions, and reduced or no participation fees. This makes innovation reachable even for those without a large compliance budget.
Testing in real-world conditions
In addition to the sandbox, the AI Act sets out rules for testing in real-world conditions outside a sandbox (Art. 60), with safeguards such as a testing plan, informed consent and the ability to halt the test. This is a separate route, with its own conditions.
How to apply in the Netherlands
In the Netherlands the Dutch Data Protection Authority (Autoriteit Persoonsgegevens) and the Dutch Authority for Digital Infrastructure (RDI) are involved in AI supervision and setting up the sandbox. The exact application procedure and admission criteria will be published by the competent authority in the run-up to the 2 August 2026 deadline; check their official channels for the current application form and conditions.
Read more: AI Act: timeline of obligations. Take the scan.
Sources
- https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Regulation (EU) 2024/1689 (AI Act), Arts 57-62: sandboxes and innovation support.
Read next
Real-world testing and regulatory sandboxes (Articles 57-60)
The AI Act offers two controlled testing routes: regulatory sandboxes under supervision (Articles 57-59) and testing in real-world conditions outside the sandbox (Article 60). Both carry strict safeguards, such as informed consent and the right to have data erased.
The authorised representative for non-EU providers (Article 22)
A provider established outside the EU must appoint a written authorised representative in the Union before placing a high-risk AI system on the market. Article 22 makes that person the European point of contact for authorities, with its own duties and power to end the mandate.
The AI Office: role, tasks and enforcement powers
The AI Office within the European Commission coordinates implementation of the AI Act and is the exclusive supervisor of GPAI models. It draws up codes of practice, conducts investigations and can have fines imposed on model providers.