EU opens AI regulatory sandboxes December 1 under common rules
Brussels published implementing rules governing AI regulatory sandboxes, with the effect date set at 1 December 2026.
· Last verified: 13 Nov 2026 (EUR-Lex (Official Journal of the European Union), EU Law Live, TechTimes)

Summary
- Brussels published implementing rules governing AI regulatory sandboxes, with the effect date set at 1 December 2026.
- Sandboxes let firms test AI systems under supervision before market launch, with small businesses prioritised and charged nothing for core guidance.
- Non-EU providers must appoint an authorised representative inside the bloc who can be held liable for compliance.
The latest
AI developers can enter supervised testing environments across the European Union from 1 December under common rules published in the Official Journal on 11 November 2026. Commission Implementing Regulation (EU) 2026/2487, adopted a day earlier, sets out how AI regulatory sandboxes operate under Article 58(1) of the AI Act. The Commission argues shared rules are needed to prevent fragmentation and secure consistent application.
Details
- The legal basis: The regulation implements Regulation (EU) 2024/1689, the AI Act, and rests on Article 58(1). Article 10 sets entry into force on the twentieth day after publication, with EUR-Lex listing the date of effect as 1 December 2026.
- What a sandbox is: The text defines sandboxes as controlled frameworks set up by competent authorities to develop, train, validate and test innovative AI systems, including in real-world conditions, under regulatory supervision and for a limited period before market placement.
- Who qualifies: Article 3(2) sets cumulative criteria: an applicant must be a provider, prospective provider or authorised representative, and the system must either not yet be on the market, be already marketed but facing substantial modification, or have become high-risk through its use.
- How entrants are picked: Article 3(3) lists selection criteria including need to participate, regulatory learning potential, public interest contribution, maturity and innovative character. Article 3(6) requires transparent, fair and equal selection with priority for SMEs, including startups.
- The timeline: Authorities decide within three months of a complete submission or of a call's closing date, according to TechTimes. Extension requests must be filed at least 30 calendar days before the planned end date, and plans can be amended by mutual agreement.
- The sandbox plan: Article 5(2) requires applicant details, an estimated schedule and end date, objectives and roles, applicable legal requirements, whether personal data is processed and whether the data protection authority must be involved, risk safeguards, incident procedures and any real-world testing plan.
- The outputs: Participants receive written proof of completed activities and an exit report, delivered within 45 working days of the end of participation under Article 6(5). Both are recognised by market surveillance authorities and notified bodies across all member states.
- The limit: Neither document counts as a declaration of conformity, and standard conformity assessment remains mandatory. Publication on the EU platform requires consent of both participant and authority; without it, only non-confidential main findings are published, TechTimes reported.
- The cost: Article 2(3) makes basic guidance and supervision free for SMEs and startups, with exceptional costs for other agreed services recoverable on a fair, transparent and proportionate basis with prior notice and an itemised invoice. Other companies can be charged in full.
- Foreign providers: Article 2(5) requires providers established outside the Union to appoint an authorised representative inside the bloc who can be held legally liable for compliance — the gateway condition for any Middle East firm seeking to test an AI system in a European sandbox.
Background
Recital 6 says member states should establish sandboxes by 2 August 2027. The AI Act's original deadline was 2 August 2026, pushed back by the digital package regulation (EU) 2026/1744 published on 24 July.
Between the lines
The deadline slipped a full year for a reason: when Parliament and Council reached their provisional agreement in May 2026, only one sandbox was operational across 27 member states, according to TechTimes. That gap explains why the European AI Office may run a Union-level sandbox itself, prioritising SMEs and startups — a fallback if national authorities again miss the date.
What's next
Watch for fee schedules, which authorities must publish when applications open, and for the first national sandboxes to launch ahead of the 2 August 2027 deadline. Data protection authorities supervise personal data processing inside them, under Recital 11.
Source: EUR-Lex (Official Journal of the European Union), EU Law Live, TechTimes