The Distributed Enforcement Gamble

Ireland’s September 2025 decision to designate 15 specialised national competent authorities for EU AI Act enforcement represents an ambitious attempt at regulatory coordination—but it also creates a coordination problem that could unravel before the August 2026 compliance deadline.

The model spreads enforcement responsibility across 15 distinct bodies, with a new National AI Office (launching 2026) tasked with coordinating supervision while hosting an innovation sandbox. On paper, this specialisation makes sense: different authorities handling high-risk AI in healthcare, employment, criminal justice, and financial services respectively. But with just 16 months until the first deadline, divergent interpretations and enforcement priorities could leave enterprises confused about which authority has final say.

Why This Matters Now

The EU AI Act’s compliance timeline already fractured during April’s Digital Omnibus trilogue negotiations. Deadlines now sit at December 27, 2027 for standalone high-risk systems and August 2, 2028 for AI embedded in regulated products—but Ireland’s August 2026 initial implementation still stands. This creates a compressed window where Ireland’s 15 authorities must establish consistent guidance before broader EU enforcement begins.

The real risk: without tight coordination mechanisms, different authorities could issue conflicting guidance on borderline cases. Is an AI resume-screener system “high-risk employment AI” (one authority’s jurisdiction) or “standalone general-purpose AI” (another’s)? Irish enterprises need clarity before investing in compliance infrastructure.

The Sandbox Problem

Ireland’s innovation sandbox—housed within the National AI Office—is meant to be a pressure valve. In theory, businesses can test high-risk systems in a controlled environment and get regulatory guidance before full deployment. But a 15-authority model risks making the sandbox itself fragmented. If your AI touches multiple risk domains, which authority’s sandbox rules apply?

Compare this to the EU’s broader struggle: member states are racing to establish their own regulatory sandboxes (as reported by regulatory coordination bodies), creating a patchwork that confuses multinational builders. Ireland, as EU presidency holder from June 30, 2026, will inherit responsibility for unifying this chaos at the exact moment its own 15-authority model hits critical testing.

Practical Implications for Builders

If you’re building high-risk AI systems in Ireland, you now face an uncomfortable choice:

  1. Wait for National AI Office guidance (late 2026) and risk missing the August deadline
  2. Engage directly with your sector’s designated authority and risk misalignment with others
  3. Over-comply across all 15 authority standards—expensive but safer

Option 3 is likely what large enterprises will choose, effectively creating a de facto “strictest interpretation” compliance floor. Smaller builders and startups won’t have that luxury.

Open Questions

  • Will the National AI Office publish coordinated guidance by Q3 2026, or will authorities diverge?
  • How will the August 2026 implementation deadline interact with the December 2027 postponement for standalone systems?
  • Will Ireland’s EU presidency (June-December 2026) give it leverage to harmonise authority positions before August enforcement kicks in?
  • Is a 15-authority model scalable when the EU AI Act expands beyond the initial high-risk categories?

The Verdict

Ireland’s distributed model is architecturally sound for a mature regulatory environment. But with 16 months to implementation and an already-splintered EU compliance timeline, coordination will make or break it. Watch for the National AI Office’s first coordinated guidance document—that timing will signal whether this ambitious experiment can deliver coherent enforcement, or whether it becomes a cautionary tale about the limits of distributed regulation.

Source: artificialintelligenceact.eu