EU AI Act regulatory-sandbox deadline postponed to August 2 2027 — member states get an extra year to stand up the experimentation infrastructure
The May 7 EU AI Act omnibus agreement postponed the deadline for member-state AI regulatory sandboxes to August 2, 2027. The original deadline was August 2 2026. The extension reflects member-state difficulty standing up the sandbox infrastructure on the original timeline and is the most concrete operational consequence of the omnibus simplification package.
The sandbox mechanism is the AI Act's intended innovation safety valve — a pathway for high-risk AI systems to be developed and tested under regulator supervision before full compliance is required. The original schedule had every EU member state operating at least one sandbox by August 2026. By May 2026 only a handful of member states (Spain, Estonia, Finland) had the infrastructure live; the rest were behind. The deadline extension acknowledges operational reality.
The downstream effect on AI startups in the EU is mixed. Sandbox access has been one of the planned counterweights to the high compliance cost the AI Act imposes on smaller developers. Pushing the universal-availability deadline out by a year means that startups in member states without an operational sandbox face the full compliance regime longer than originally planned — until either their member state stands up a sandbox or they relocate development to a member state that has one.
Consilium — AI Council and Parliament agree to simplify and streamline rules → · Latham & Watkins — AI Act Update: EU Resolves to Change Rules and Extend Deadlines → · AI Act EU — EU Artificial Intelligence Act developments →