The Sectoral Standoff That Could Unravel EU AI Compliance

When EU negotiators walked away from the Digital Omnibus trilogue on April 28 after just 12 hours, the flashpoint wasn’t regulatory scope or enforcement timelines—it was a technical architectural question that could fundamentally reshape how AI safety is overseen across the bloc.

The dispute centers on Annex I conformity assessment: whether AI systems embedded as safety components in machinery, medical devices, toys, and diagnostics should continue through sectoral regulators (Section B) or shift to a centralised European AI Board track (Section A). This isn’t semantic—it’s the difference between distributed oversight and unified governance.

What Failed Last Time

The Council and Parliament couldn’t agree whether Section A products (those requiring independent conformity assessment) should move into Section B’s sectoral framework. The implication: if moved, the European AI Board’s authority contracts, and responsibility fragments across medical device authorities, machinery safety bodies, and toy regulators. The Parliament generally favours centralised scrutiny; the Council prefers sectoral expertise.

For European enterprises building AI in regulated products, this uncertainty is costly. If conformity assessment architecture shifts, your certification pathway changes. A medical device manufacturer integrating AI diagnostics today may face a different approval regime by August 2026.

The Timeline Pressure

The Cyprus Presidency has until June 30 to close this file. A follow-up political trilogue is scheduled for approximately May 13—just two weeks after collapse. If talks fail again, negotiations pass to the Irish Presidency (beginning July 1), introducing another vector of uncertainty.

Here’s the critical detail: the August 2, 2026 deadline for stand-alone Annex III high-risk systems remains legally binding, regardless of whether Annex I conformity architecture is resolved. This creates asymmetrical pressure. Stand-alone high-risk builders have a fixed deadline; embedded AI builders face both deadline uncertainty and pathway uncertainty.

What Hasn’t Changed

GPAI obligations (Articles 50-55) were not in dispute and remain on schedule. If you’re operating large language models or foundation models under European governance, treat the Omnibus collapse as noise, not signal. Your compliance roadmap is unaffected.

Practical Implications for Irish and European Builders

If you’re integrating AI into medical devices, machinery, or diagnostics:

  • Document your current conformity pathway and monitor May 13 outcomes closely
  • Plan for two scenarios: Section B sectoral oversight (current path) and potential Section A centralisation
  • Engage with your national competent authority to understand their preferred approach—sectoral bodies have political voice in Council negotiations
  • For August 2026 compliance: treat this deadline as fixed regardless of Annex I resolution; assume you’ll navigate the existing sectoral framework

For standalone high-risk AI builders, the deadline is non-negotiable. But for embedded safety-critical systems, architectural clarity matters more than timeline extension.

The Irish Question

With Ireland assuming the Presidency on July 1, there’s a window where this file could either close under Cyprus or reopen under Dublin. Ireland’s approach to centralised versus distributed AI oversight could influence whether the next trilogue (if needed) pushes toward Board authority or sectoral pragmatism. Watch for Ireland’s position statements in late May.

The May 13 trilogue will determine whether European builders face one August 2026 deadline or a splintered compliance landscape.

Source: artificialintelligenceact.eu