European Union flags in front of the Berlaymont building, headquarters of the European Commission in Brussels
© European Union, 2026, licensed under CC BY 4.0 (EC Audiovisual Service, P-069848/00-06); Berlaymont: Lucien De Vestel, Jean Gilson, A&J Polak; reno. Berlaymont 2000

News

EU sets common rules for AI regulatory sandboxes

Implementing Regulation (EU) 2026/2487, adopted on 10.11.2026 and published on 11.11.2026, sets common EU rules for supervised testing of AI systems before market entry, with priority access for SMEs and start-ups.

By Nova 5 min read

Key takeaways

  • Implementing Regulation (EU) 2026/2487 was adopted on 10.11.2026, published in the Official Journal on 11.11.2026, and enters into force on 01.12.2026.
  • Member States must have at least one national AI regulatory sandbox operational by 02.08.2027.
  • SMEs, including start-ups, and public authorities receive priority access and, as a rule, free regulatory guidance, advice and supervision, subject to possible exceptional costs.
  • Written proof and the exit report document sandbox work. They do not have the legal effect of a declaration of conformity.
  • The CNPD already runs the Sandkëscht data protection sandbox in Luxembourg, and the official RE.M.I. plenary programme on 12.11.2026 in Belval included a working group on the governance of AI agents.
On this page
  1. What does Implementing Regulation 2026/2487 change?
  2. Who can use an AI regulatory sandbox for free?
  3. Do exit reports authorise a product launch?
  4. What does the new sandbox regulation mean for Luxembourg?

The European Commission adopted Implementing Regulation (EU) 2026/2487 on 10.11.2026, published in the Official Journal on 11.11.2026, setting common rules for AI regulatory sandboxes in which firms can develop and test systems under supervision before they enter the market.

What does Implementing Regulation 2026/2487 change?

The act sets out the detailed arrangements that Article 58 of the AI Act, Regulation (EU) 2024/1689, requires the Commission to adopt, so that sandboxes do not fragment across the Union. It applies to all AI regulatory sandboxes established under Article 57 and enters into force on 01.12.2026, the twentieth day after publication in the Official Journal.

Article 57 of the AI Act, in the consolidated version of 27.07.2026 published by the Commission's AI Act Service Desk, requires Member States to ensure that their competent authorities "establish at least one AI regulatory sandbox at national level, which shall be operational by 2 August 2027." The sandbox may be set up jointly with other Member States, or a Member State may take part in an existing sandbox if this gives equivalent national coverage. The AI Office may also establish a Union-level sandbox for certain systems, with priority access for SMEs, including start-ups, and small mid-cap companies.

These sandboxes are a controlled framework for the development, training, validation and testing of innovative AI systems, including in real-world conditions where appropriate, under a sandbox plan and for a limited time before they are placed on the market.

Who can use an AI regulatory sandbox for free?

Access is not automatically free for every participant. For SMEs, including start-ups, access and participation "shall be prioritised and free of charge" for regulatory guidance, advice and supervision concerning the AI Act and other law supervised in the sandbox, including supervision of testing in real-world conditions. Authorities may still recover exceptional costs for other services agreed in the sandbox plan, in a fair, transparent and proportionate way. Prospective participants must be told about that possibility at least before participation starts, and any invoice must be sufficiently detailed.

Public authorities, bodies and agencies benefit from the same pricing regime. For undertakings that are not SMEs (start-ups fall within the SME category here), authorities may charge a fee equal to the full costs incurred and must publish the fee list when they open the application phase. The authority must inform the applicant of the selection outcome no later than three months after a complete application or after the deadline of a call. If selected, the parties agree a sandbox plan without undue delay, covering the timeline, objectives, legal requirements supervised, and any testing in real-world conditions.

Weekly, Thursday morning, Central European Time.

The Thinking briefing

One email a week: what changed in European business software, which prices moved, and which comparison we published. No sales mail.

Do exit reports authorise a product launch?

Participation is not an approval to place a system on the market. After the project ends, the authority issues written proof of the activities carried out. That proof "shall not have the same status or legal effect as a declaration of conformity within the meaning of Article 47" of the AI Act. An exit report must follow within a reasonable period, and no later than 45 working days after participation ends. Under Article 57, both documents "shall be taken positively into account by market surveillance authorities and notified bodies, with a view to accelerating conformity assessment procedures to a reasonable extent."

What does the new sandbox regulation mean for Luxembourg?

Luxembourg already operates a regulatory sandbox. The National Commission for Data Protection (CNPD) runs Sandkëscht, which it describes as "an environment dedicated to testing and understanding the legal implications of new technologies and/or new uses of data, particularly in the field of artificial intelligence." The programme focuses on personal data protection. Any establishment registered in Luxembourg may apply, whatever its size, sector or status, if the project raises a data protection issue and offers benefits for individuals or society. Participation generally lasts from 9 to 18 months, does not exempt participants from their legal obligations, and does not include IT infrastructure or data from the CNPD.

The CNPD page for submitting a project, last updated on 15.01.2026, states that the pilot phase is closed and invites organisations to submit a project to the artificial intelligence regulatory sandbox in August 2026. The pages reviewed on 12.11.2026 do not state whether Sandkëscht will be Luxembourg's national sandbox under Article 57 of the AI Act.

On the same date, the CNPD published the programme of the 4th plenary of RE.M.I. (Regulation Meets Innovation), created by the CNPD and the AI Factory. The meeting was scheduled for 12.11.2026 from 08:30 to 12:00 at the Terres Rouges Building in Belval, in Esch-sur-Alzette. Opening remarks were listed for David Benhamou, the CNPD's regulatory sandbox lead. The official agenda included the launch of an AI Agent Governance Working Group, presented by Dr Alessio Buscemi of LIST and Giuseppe Tarantino of Luxinnovation.

Paved plaza in the Belval district of Esch-sur-Alzette with contemporary buildings, people seated on a bench beside a reflecting pool, and a preserved industrial wall under a clear blue sky
Street life in the modern Belval district of Esch-sur-Alzette, where the University of Luxembourg is located, September 2022 | Photo: Sabino Parente/Shutterstock.com

For Luxembourg SMEs seeking public support for an AI project, the steps for SME Packages AI remain a separate national funding route. Teams that process personal data in everyday tools should still review their GDPR duties for AI email before they test a system, in a sandbox or outside one. The implementing regulation applies directly in all Member States from 01.12.2026. National sandboxes under the AI Act must be operational by 02.08.2027.

Questions and answers

Is an AI regulatory sandbox free for every company?
No. SMEs, including start-ups, and public authorities benefit from a free-of-charge regime for regulatory guidance, advice and supervision, with possible exceptional costs for other services agreed in the sandbox plan. Other undertakings may be charged the full costs.
Does an exit report allow a system to be placed on the market?
No. Written proof and the exit report document the activities carried out in the sandbox. They do not have the same legal effect as a declaration of conformity under Article 47 of the AI Act.

Sources

Every price and fact above was checked against these pages on the date shown.

  1. [1] Commission Implementing Regulation (EU) 2026/2487 of 10 November 2026 (Official Journal L series, 11.11.2026) accessed 2026-11-12
  2. [2] Commission Implementing Regulation (EU) 2026/2487, Official Journal PDF accessed 2026-11-12
  3. [3] Article 57: AI regulatory sandboxes, AI Act Service Desk (consolidated text as at 27 July 2026) accessed 2026-11-12
  4. [4] Article 58: Detailed arrangements for, and functioning of, AI regulatory sandboxes, AI Act Service Desk accessed 2026-11-12
  5. [5] RE.M.I. Initiative, 4th Plenary Session, CNPD (English) accessed 2026-11-12
  6. [6] Initiative RE.M.I., 4e session plénière, CNPD (français) accessed 2026-11-12
  7. [7] Sandkëscht, Regulatory Sandbox, CNPD accessed 2026-11-12
  8. [8] Sandkëscht, frequently asked questions, CNPD accessed 2026-11-12
  9. [9] Sandkëscht, criteria for participation, CNPD accessed 2026-11-12
  10. [10] Sandkëscht, submit a project, CNPD accessed 2026-11-12
  11. [11] RE.M.I., Regulation Meets Innovation, CNPD accessed 2026-11-12
Topics ai-actsandboxluxembourgcnpd