Proposed Amendments to the EU AI Act: Embedding Quality by Design

Peter Haenni

• 2 min read

The purpose of the proposed amendments is to enhance the EU AI Act by integrating the Quality by Design (QbD) principle, ensuring that high-risk AI systems are developed not only with safety and risk mitigation in mind, but also with robustness, reliability and consistent high quality built-in from the very start.

 Key Amendments

  1. New Recital The proposed recital emphasizes that AI systems placed on the EU market must meet quality objectives from the outset. It advocates for the application of the Quality by Design (QbD) principle throughout the AI system lifecycle, fostering continuous improvement, transparency and resilience.
  2. Amendment to Article 9 (Risk Management System) A new provision clarifies that risk management must align with quality management and QbD objectives, promoting a balanced and holistic approach to AI system development and operation.
  3. Amendment to Article 10 (Data and Data Governance) Strengthening the data governance requirements to ensure that data quality and representative datasets align with the quality objectives defined under the QbD approach.
  4. Amendment to Article 15 (Accuracy, Robustness and Cybersecurity) The emphasis shifts from purely risk-based approaches to include proactive quality objectives for ensuring the robustness, accuracy and cybersecurity of AI systems.
  5. Creation of New Article 17a (Quality by Design Requirements) This article lays out the foundational principles of QbD for AI systems, including:
      • Early identification and integration of measurable quality objectives
      • Detailed planning for data quality, model performance and system robustness
      • Continuous monitoring, feedback and improvement post-deployment.

Providers will be required to document their application of QbD principles, ensuring transparency and accountability.

6. Harmonized Standards and Common Specifications The Commission is empowered to adopt harmonized standards and common specifications to guide the practical application of QbD in AI systems, supporting industry-wide consistency and high-quality AI development.

Rationale for Amendments

The EU AI Act as it stands rightly focuses on risk mitigation, but to truly ensure that AI systems in Europe are reliable, transparent and high-performing, it is necessary to integrate quality management principles from the outset. These amendments aim to:

  • Promote proactive quality assurance in the design and development of AI systems
  • Foster long-term AI system performance and sustainability across the AI lifecycle
  • Align with global best practices from other industries, such as pharmaceuticals and manufacturing, where Quality by Design ensures that quality is systematically built into the process, rather than treated as an afterthought.

Embedding QbD in the EU AI Act will ultimately contribute to trusted, resilient AI systems that meet the needs of both users and society, reducing the risk of failures or unforeseen negative impacts while fostering innovation.

 Conclusion

By adopting these amendments, the EU AI Act would go beyond risk mitigation, setting a global standard for quality in AI development. This comprehensive, quality-first approach will not only strengthen trust in AI systems but also empower developers to produce AI that meets both ethical and technical excellence standards. With Quality by Design as a core principle, Europe can lead the way in high-quality, sustainable AI innovation.