EU AI Act Horizontal Approach

The EU Artificial Intelligence Act adopts a horizontal regulatory approach, enacting harmonized rules across all economic sectors and domains within the European single market, rather than formulating separate, siloed regulations for each vertical industry (such as finance, healthcare, or transportation).

Horizontal vs. Sector-Specific Governance

Sector-Specific (Vertical) Approach           Horizontal (Cross-Sector) Approach
┌───────────┐ ┌───────────┐ ┌───────────┐     ┌─────────────────────────────────────┐
│  Health   │ │ Transport │ │  Banking  │     │           EU AI Act Scope           │
│ AI Rules  │ │ AI Rules  │ │ AI Rules  │     │ (Harmonized Risk-Based Obligations) │
└─────┬─────┘ └─────┬─────┘ └─────┬─────┘     └──────────────────┬──────────────────┘
      ▼             ▼             ▼                              ▼
 [Fragmented Single Market & Arbitrage]       [Cross-Cutting Application: Health, HR,
                                               Critical Infra, Banking, Public Sector]

Under a vertical regime, individual sector regulators develop bespoke rules. While vertical rules accommodate domain nuances, they create regulatory fragmentation, overlapping authority, and loopholes for general-purpose technologies that traverse multiple sectors.

In contrast, the European Commission structured the AI Act under the New Legislative Framework (NLF) for product safety. The Act establishes uniform, horizontal safety and fundamental rights baselines that apply equally to an AI system whether deployed in medical diagnostics, employment screening, law enforcement, or industrial automation.

Strategic Drivers for a Horizontal Model

Single Market Harmonization: Divergent national or sector-level AI safety mandates would create internal barriers to trade, distort competition, and prevent European AI vendors from scaling across member states.
Mitigating Regulatory Arbitrage: General-purpose AI models and shared machine learning components could exploit jurisdictional and sectoral seams if compliance requirements varied by industry.
Risk-Tiered Calibration: Instead of differentiating by industry sector, the regulation differentiates by use-case risk category (unacceptable, high-risk, transparency-specific, minimal). A recruitment algorithm is classified as high-risk regardless of whether the employer operates in logistics, banking, or agriculture.

References

Proposal for a Regulation laying down harmonised rules on Artificial Intelligence COM(2021) 206 final - Explanatory Memorandum detailing the justification for a horizontal, single-market instrument.
Regulation (EU) 2024/1689 (Artificial Intelligence Act) - Final legislative text confirming the cross-sectoral product safety architecture.
EU AI Regulatory Policy Timeline - Context on the shift toward horizontal harmonization in 2021.
Understanding the Foundations of the EU AI Act - Module 1 - Source lecture module explaining the rationale of the horizontal approach.

Review cards

Flashcard What is the New Legislative Framework (NLF), and how does the EU AI Act use it?

The NLF is the EU's standard model for product-safety laws: essential requirements, conformity assessment, CE marking and market surveillance. The AI Act is built on it, so a high-risk AI system is treated like a regulated product that must show conformity before it reaches the market.

Flashcard How does the horizontal model of the EU AI Act address the issue of regulatory arbitrage?

By establishing uniform, cross-cutting regulations that apply to AI systems regardless of their sector of deployment, thus preventing general-purpose AI models from exploiting jurisdictional and sectoral differences in compliance requirements.

Quiz Which of the following best describes the risk-tiered calibration in the EU AI Act?

Risk is differentiated by industry sector.
Risk is categorized into unacceptable, high-risk, transparency-specific, and minimal.
Risk is only assessed for high-risk sectors like healthcare and finance.
Risk is set by the size or architecture of the AI model.

The regulation differentiates risk by use-case, not by industry sector, ensuring a consistent approach across various applications.