There's a push for machinery and medical device manufacturers to get exemptions from the EU AI Act. The argument is that sector-specific regulations should handle AI in these areas because the AI Act is too broad and burdensome. Germany and the centre-right in Parliament support this view. It seems logical: why make a CNC machine with vision systems comply with the AI Act when the Machinery Regulation already covers safety?
Here's why that logic doesn't hold up.
Why the Horizontal Approach Matters
Sector-specific laws weren't designed for AI. They focus on mechanical failure and electrical hazards, not on AI-specific issues like training data bias or model drift. The Medical Device Regulation (MDR) requires clinical evaluation but doesn't address dataset representativeness or fairness metrics. The Machinery Regulation mandates risk assessments for physical hazards but doesn't cover predictive maintenance algorithms that misclassify due to different training environments.
You can comply with sectoral laws and still deploy AI systems that disadvantage certain populations or create new workplace risks. The AI Act's horizontal approach fills this gap by setting baseline requirements for AI-specific risks across all domains, like data governance under Article 10 and transparency requirements in Article 13. These complement, not duplicate, sectoral rules.
Insights from the Trilogue Breakdown
EU legislators couldn't agree on delaying the AI Act, with machinery and medical devices as key issues. The centre-right wanted these products to follow sectoral law instead of the AI Act, but several member states and the centre-left disagreed.
This isn't just about regulatory burden; it's about regulatory structure. Germany's Social Democrats warned that the proposed changes could create inconsistent rules and benefit foreign companies over European ones. Allowing machinery manufacturers to opt out of AI Act requirements gives an advantage to those who don't have to prove data quality or maintain monitoring systems.
The new rules for high-risk AI applications are set to apply this August. The breakdown in negotiations tightens the timeline but also reinforces the importance of the horizontal approach.
Steps to Take Now
Don't wait for exemptions that might not come. Start integrating AI Act requirements with your existing sectoral compliance.
For AI in medical devices, expand your clinical evaluation under MDR. Document how you validated the algorithm's performance across demographic groups and monitor for shifts between training data and clinical populations. Extend post-market surveillance to track AI-specific failure modes like prediction errors.
For machinery with AI components, include algorithmic risks in your risk assessment under the Machinery Regulation. Document training data conditions and define acceptable adaptation boundaries.
The AI Act adds a layer to sectoral requirements. Your compliance program should address both, with the integration point being your risk assessment. Article 9 requires a risk management system throughout the AI system lifecycle. This should reference your MDR clinical evaluation or Machinery Regulation safety assessment and extend to cover data quality and model governance.
Practically, update your design controls to include AI-specific validation steps. Expand technical documentation to capture training data provenance and fairness metrics. Train your teams to recognize when an AI component triggers high-risk classification under Annex III.
The Role of Sector-Specific Expertise
Domain expertise is crucial for AI governance, and sectoral regulators bring that expertise. Medical device notified bodies understand clinical risk, and machinery safety assessors know how industrial equipment fails. This knowledge is essential for effective AI governance.
The solution isn't to exempt these sectors from the AI Act but to ensure sectoral regulators develop AI competence and coordinate with the AI Office. Article 77 provides for cooperation between the AI Office and sectoral authorities.
Engage with both. When preparing technical documentation for a high-risk AI system, write for both the notified body and the market surveillance authority. Integrate both perspectives.
The machinery and medical device industries have valid concerns about compliance complexity. The answer isn't a carve-out but better coordination between horizontal and vertical regulation, clearer guidance from the AI Office, and compliance programs that treat the AI Act as an extension of existing safety culture.
The horizontal approach is here to stay. Build your governance program accordingly.
