DEV Community

News
News

Posted on

EU adopts common rules for AI regulatory sandboxes: Implementing Regulation 2026/2487 applies from December 1

Infographic cover

๐Ÿ“… November 11, 2026 โ€ข Source: Official Journal of the EU (EUR-Lex)

The European Commission adopted Implementing Regulation (EU) 2026/2487 on November 10, 2026. It was published in the Official Journal of the EU on November 11, 2026 and applies from December 1, 2026. Issued under Article 58 of the AI Act (Regulation (EU) 2024/1689), the regulation lays down common rules for the establishment, development, implementation, operation and supervision of AI regulatory sandboxes across the Union.

The rules apply to every AI regulatory sandbox run by member state competent authorities under Article 57 of the AI Act, aiming to avoid fragmentation and guarantee consistent implementation. Each member state must have at least one national sandbox operational by August 2, 2027, a deadline the Digital Omnibus amendment moved from August 2, 2026. Participants receive written proof of completed activities and an exit report, but neither document carries conformity status โ€” sandbox tests earn no compliance clearance.

Market surveillance authorities and notified bodies in every member state must take both documents into account as a record of the work done. The exit report must arrive within a reasonable time and no later than 45 working days after participation ends, covering the main regulatory issues examined, how they were resolved, recommended practices and benchmarks, and any serious incidents. Participants remain liable for third-party damage and EU law compliance, but the AI Act bars administrative fines when a participant follows the sandbox plan and authority guidance in good faith.

Authorities can suspend participation if plan conditions are breached, if significant risks cannot be reduced, or at the participant request, and the Commission AI Office must be informed without undue delay. Providers and prospective providers can apply, with priority access plus free core regulatory guidance and supervision for SMEs โ€” authorities must reply within 3 months of a complete application or the call deadline. Providers without an EU office must pay the participation fee and name an authorized representative in the Union.

Publication of results on the EU platform requires explicit agreement from both sides, or authorities publish only non-confidential key findings. Sandbox plans list the AI Act obligations to be examined, with Article 15 cybersecurity duties covering data poisoning and adversarial examples. High-risk AI obligations follow from December 2, 2027 for stand-alone systems and August 2, 2028 for AI in regulated products. In practice, companies test AI under supervision before market placement, the exit report is a recognized record across the EU, and SMEs get the cheapest entry.

Sources: data.europa.eu - ai-act-service-desk.ec.europa.eu - theworldofai.org

Key figures

IMPACT checklist


Sources:


Continue the conversation: https://t.me/AIAutomationArchitecture ยท Bluesky

AI Automation

Top comments (0)