Digital Omnibus on AI: Subsidiarity and Proportionality Assessment

When the European Commission published COM(2025)836 — the Digital Omnibus on Artificial Intelligence — it triggered the EU's standard subsidiarity and proportionality review process. National parliaments exercised their right under Protocol No. 2 to the Treaty on the Functioning of the European Union to scrutinise whether the proposal respected the limits of EU competence, while the Council's legal service and the European Parliament's committees conducted parallel assessments. This article explains what subsidiarity and proportionality mean in EU law, how they were applied to the Digital Omnibus AI proposal, and what the outcome means for the final text of the EU AI Act's Article 6 amendments.

What Subsidiarity and Proportionality Mean in EU Law

Subsidiarity — Article 5(3) TEU. The principle of subsidiarity, codified in Article 5(3) of the Treaty on European Union, provides that in areas of shared competence the EU should act only if and insofar as the objectives of the proposed action cannot be sufficiently achieved by member states acting alone, but can better be achieved at Union level by reason of the scale or effects of the proposed action. In practice, the Commission must demonstrate a cross-border dimension to the problem and show that uncoordinated national action would produce fragmentation of the internal market.

For AI regulation, the subsidiarity case is generally straightforward: AI systems are developed and deployed across borders, AI supply chains are pan-European and global, and divergent national AI rules would directly impede the free movement of AI-enabled goods and services. The EU AI Act's original subsidiarity justification was accepted without significant objection. However, the Digital Omnibus's proposed amendments — which in several respects devolve detail to member state competent authorities and national sandboxes — required a more nuanced subsidiarity analysis.

Proportionality — Article 5(4) TEU. The principle of proportionality, codified in Article 5(4) TEU, requires that the content and form of Union action shall not exceed what is necessary to achieve the objectives of the Treaties. For legislative proposals, the Commission's impact assessment must demonstrate that the chosen regulatory approach imposes no more burden than necessary to achieve the identified policy objective, and that less restrictive alternatives have been considered.

Proportionality is the sharper edge of the two principles for the Digital Omnibus: member state governments and business associations had raised sustained concerns that certain AI Act obligations — in particular the conformity assessment procedures under Article 43 and the GPAI model obligations under Articles 51–55 — imposed costs disproportionate to the actual risk reduction achieved. COM(2025)836 is, at its core, a proportionality correction.

Member State Assessments of the Digital Omnibus AI Proposal

Following publication of COM(2025)836, national parliaments submitted reasoned opinions under the "yellow card" procedure of Protocol No. 2. No yellow card threshold (one-third of parliamentary votes) was reached, meaning the Commission was not required to review the proposal. However, the reasoned opinions submitted by several chambers — including the German Bundesrat, the Dutch Eerste Kamer, and the Austrian Nationalrat — identified substantive proportionality concerns that fed into the Council's negotiating mandate.

German Bundesrat concerns. The German Bundesrat, which represents the Länder governments in the German federal structure, raised concerns that COM(2025)836's revised Article 6 classification mechanism — the presumption of high-risk — could create uncertainty for manufacturers in precision engineering and automotive supply chains where AI-assisted quality control systems sit close to the Annex III boundary. The Bundesrat argued that the rebuttal criteria to be published by the AI Office under the revised Article 6 should be published before the amended regulation enters into force, to allow advance planning.

Dutch Eerste Kamer concerns. The Dutch chamber focused on over-harmonisation risk: that by legislating in detail on SME derogation conditions, COM(2025)836 would prevent member states from providing more protective national rules for workers and consumers affected by AI systems operated by SME employers. This concern was framed as a subsidiarity issue — the argument being that SME support measures are better calibrated at member state level given differences in national labour markets and social welfare systems.

Austrian Nationalrat concerns. The Austrian parliament expressed proportionality concerns about the revised GPAI tier thresholds in the proposed Article 51, arguing that the revised compute thresholds were derived from current hardware economics and would require rapid revision as AI compute costs fall, creating legal uncertainty.

How the Assessment Shaped Article 6 Amendments

The Council's working group on AI (WGAI) incorporated feedback from the national parliament opinions and from member state technical experts into the revised Council mandate for trilogue. The key modifications to COM(2025)836's Article 6 proposal were as follows.

First, the mandate requires that the AI Office publish the rebuttal criteria for the Article 6 presumption mechanism at least six months before the amended regulation's application date. This directly addresses the German Bundesrat's concern about advance planning, and gives providers of Annex III systems a defined window to assess whether their systems qualify for rebuttal.

Second, the Council mandate includes a review clause requiring the Commission to reassess the Annex III scope and the Article 6 criteria within three years of application, having regard to technological developments and market surveillance data. This dynamic element responds to the Austrian concern about technological obsolescence of fixed thresholds.

Third, on the Dutch over-harmonisation point, the mandate preserves member state competence to introduce complementary national rules protecting workers from AI-assisted employment decisions, provided such rules do not create barriers to the free movement of AI systems. This compromise acknowledges the subsidiarity concern without undermining the internal market objective.

Implications for Businesses

The subsidiarity and proportionality process, though often perceived as a procedural formality, produced substantive changes to the Digital Omnibus text that directly affect business planning.

The requirement for AI Office publication of Article 6 rebuttal criteria before the amended regulation applies gives providers a clearer compliance pathway than the original Commission proposal. Companies developing AI systems in Annex III-adjacent categories — including HR screening tools, educational assessment systems, and manufacturing quality control applications — should monitor AI Office publications closely once the amending regulation enters into force.

The three-year review clause introduces planned legislative uncertainty: Annex III scope and the Article 6 criteria may be revised within the compliance cycle of systems currently in development. Product roadmap and conformity assessment investments should factor in this review risk, potentially by building modularity into compliance documentation structures.

The preservation of member state worker protection rules means that companies operating in multiple EU jurisdictions may face national-level AI-in-employment rules layered on top of the EU AI Act baseline. Labour law counsel in each key operating jurisdiction should be engaged as part of AI deployment planning for human resources applications.

Conclusion

The subsidiarity and proportionality review of COM(2025)836 demonstrates that the EU's constitutional principles of shared competence governance are live mechanisms, not formalities. Member state parliaments identified real concerns — advance publication of criteria, technological obsolescence, worker protection — that left tangible marks on the Council's negotiating mandate. The resulting trilogue position is more operationally workable than the Commission's original proposal, but introduces new planning variables that compliance professionals must incorporate into their AI governance programmes.

This article is informational only and does not constitute legal advice. Consult qualified legal counsel for advice specific to your organisation.

Frequently Asked Questions

What is the yellow card procedure and did it apply to COM(2025)836?

The yellow card procedure allows national parliaments to collectively challenge a Commission proposal on subsidiarity grounds if reasoned opinions from at least one-third of parliamentary votes are received within eight weeks of the proposal's transmission. For COM(2025)836, the threshold was not reached. The reasoned opinions submitted nonetheless informed the Council's political assessment of the proportionality concerns and shaped the revised mandate, demonstrating that the yellow card procedure has influence even when the formal threshold is not met.

How does Article 5(4) TEU proportionality differ from the better regulation impact assessment?

Article 5(4) TEU proportionality is a Treaty-level legal constraint: if a measure is disproportionate, it can be challenged before the Court of Justice of the EU on grounds of Treaty infringement. The Commission's better regulation impact assessment is a procedural tool designed to help the Commission satisfy the proportionality test and choose the least restrictive effective measure. Both the legal principle and the procedural tool were engaged in the review of COM(2025)836.

Can a company rely on a member state's more favourable national implementation if the AI Act is a maximum harmonisation instrument?

The EU AI Act is a maximum harmonisation regulation in its core technical requirements for high-risk AI systems. Member states cannot impose stricter or different technical requirements on AI systems within the AI Act's scope. However, in domains where the AI Act explicitly preserves national competence — including labour law and certain data protection matters — member states may apply additional national rules. The Council mandate's preservation of worker protection rules operates within this exception.

Sources

• Protocol No. 2 on the Application of the Principles of Subsidiarity and Proportionality — EUR-Lex
• COM(2025)836 — Digital Omnibus on AI — European Commission
• EU AI Act (Regulation (EU) 2024/1689), Article 6 — EUR-Lex