Inside the EU AI Regulatory Sandbox

⚠️ IMPORTANT REGISTRATION NOTICE
(YOU HAVE TO JOIN THROUGH LUMA TO BE REGISTERED FOR THE EVENT (LINK PROVIDED).

EU AI Act Demystified: Inside the EU AI Regulatory Sandbox
Format: 75-minute briefing and live Q&A
Audience: Open to everyone, free, no prior session required
Hosted: DAIG community session

There's no single "EU sandbox" to apply to. Which one's actually relevant depends on where you, or the model you're building on, are established, and that answer is often more surprising than people expect.

Session Agenda
Part 1: What a Sandbox Actually Is (10 min)
A controlled, supervised path to test and demonstrate compliance in dialogue with a regulator, before market placement. Not an exemption, and not a compliance holiday.

Part 2: The Real Mechanics (10 min)
The documented exit report you walk away with, and why it counts as real evidence in a later conformity assessment. What Article 57(12)'s liability protection actually covers, and what it doesn't, an unfit system still faces the full Article 99 penalty regime once it reaches the market.

Part 3: Priority Isn't a Guarantee (8 min)
SME and start-up priority is real, reduced fees, priority in a transparent selection process, but entry isn't automatic. Capacity is limited, especially in early cohorts, and eligibility criteria still apply.

Part 4: Where It Actually Happens (15 min)
Ireland, the Netherlands, and Belgium, compared directly: a single coordinating authority sitting above 15 sectoral regulators, a decentralised model with one shared portal, and a country hosting the EU institutions themselves while lagging behind on its own domestic rollout. National deadline for every member state: 2 August 2027 (deferred from 2 August 2026).

Part 5: Why This Matters Even If You Never Apply (7 min)
Most people in the room would most likely be building on someone else's foundation model, not training their own, and the authority regulating that model likely isn't the one running your local sandbox.