The Sectoral Exemptions Standoff: What’s Really at Stake

The second trilogue on the EU AI Act collapsed on 28 April 2026 not over headline-grabbing issues, but over a technical question with profound implications: should AI systems embedded in products already regulated by EU sectoral legislation be exempted from the AI Act’s additional requirements?

It sounds like bureaucratic housekeeping. It isn’t. MEP Michael McNamara’s warning—that routing AI governance through sectoral legislation could prove “deregulatory rather than simplifying”—cuts to the heart of a strategic problem that could fundamentally weaken Europe’s AI safety ambitions.

The Core Problem: Fragmentation Over Integration

The proposal being debated would exempt AI systems in medical devices, industrial machinery, toys, and connected cars from AI Act requirements if they’re already covered by sector-specific EU rules. On the surface, this seems logical: why duplicate oversight?

But here’s the trap. Sectoral regulations like the Medical Devices Directive or Machinery Directive were written for a pre-AI world. They don’t contemplate the specific risks AI introduces—opacity, training data bias, model drift, adversarial robustness, or the cross-cutting governance challenges that apply across sectors.

An AI system embedded in medical imaging hardware, for example, faces safety requirements under the Medical Devices Directive. But that directive doesn’t mandate explainability standards, training data documentation, or model performance monitoring against adversarial inputs. By exempting such systems from the AI Act, the EU risks creating safety gaps that sectoral regulators simply aren’t equipped to fill.

Why This Matters for Irish Enterprises

Ireland’s position is particularly acute. With 60% of EU AI-related fines in Q1 2026 (€250 million across 50 cases) handled by Irish authorities, the country is already the enforcement epicentre for general-purpose AI (GPAI) compliance. A sectoral exemptions framework would create a two-speed compliance landscape: high scrutiny for GPAI providers, potential regulatory arbitrage for companies embedding AI in regulated products.

Irish tech firms and their subsidiaries operating high-risk AI systems face an August 2, 2026 compliance deadline. Clarity on sectoral exemptions directly affects their compliance roadmap and risk exposure.

The Practical Fallout

For AI builders and enterprises, sectoral exemptions create genuine uncertainty:

Compliance ambiguity: Is your AI system in a medical device exempted or covered? Which rules apply if sectoral and AI rules conflict?

Market fragmentation: Companies might relocate riskier AI deployments to products regulated by weaker sectoral frameworks, creating a race to the bottom.

Enforcement inconsistency: If some AI systems avoid AI Act oversight, enforcement becomes sector-dependent rather than principles-based.

The Road Ahead: Mid-May and Ireland’s Presidency

The trilogue resumes in mid-May, with Ireland assuming the EU presidency on 30 June. This timing matters. Ireland’s enforcement experience and diplomatic position could shape how the EU resolves this sectoral question.

McNamara’s framing is crucial: the goal should be simplification of overlapping rules, not exemption from safety requirements. A principled approach would establish clear AI-specific requirements (transparency, testing, monitoring) that apply regardless of sector, while avoiding duplication of established product safety processes.

Open Questions

What constitutes adequate sectoral oversight of AI risks? How will the Commission coordinate between AI and sectoral regulators? And most pressingly for August 2026: will final clarity arrive in time for compliance, or will the sectoral question remain unresolved, forcing enterprises into conservative, expensive over-compliance?

For now, European AI builders should assume the AI Act applies to their systems until exemptions are formally finalised. The burden of proof for exemption should be high.

Source: EU AI Act Trilogue Developments