Trustworthy and responsible AI beyond compliance
This recommended practice describes a governance framework for assuring AI-enabled systems. It provides guidance on how to assure that AI-enabled systems are trustworthy and managed responsibly throughout their entire life cycle.
The recommended practice is part of the UK Centre of Data Ethics and Innovation portfolio of AI assurance techniques supporting the development of trustworthy AI.
To get access to a complimentary copy of this recommended practice you need a Veracity account and a subscription to the Rules and Standards Explorer+ on Veracity which will give you access the entire database of DNV’s energy standards and recommended practices.
Key benefits of the RP for assurance of AI-enabled systems
- Beyond compliance: The RP extends beyond mere compliance. It emphasizes that AI systems should not only meet legal requirements but also perform as intended. Trustworthy AI involves more than ticking boxes; it’s about ensuring reliable outcomes and responsible use.
- Practical interpretation of legal requirements: DNV’s RP bridges the gap between the EU AI Act (which can be complex and challenging to understand) and system-specific conformity cases. It provides a practical interpretation of the legal requirements, helping stakeholders identify applicable rules and gather evidence for their compliance claims.
- Focused conformity: By tailoring the RP to AI-enabled systems, it ensures that stakeholders address the unique aspects of these systems. Unlike traditional mechanical or electric systems, AI-enabled systems evolve rapidly. The RP accounts for this dynamic nature, emphasizing a rigorous assurance methodology.
- Validity and assurance: While rubber stamps typically have a fixed validity (e.g., five years), AI-enabled systems require continuous validation. Each data point collected can impact system performance. DNV’s RP recognizes this need for ongoing assurance and adapts to the changing landscape.
- Holistic approach: Recognizing that AI quality depends on various components, DNV offers a suite of RPs covering essential building blocks: data, sensors, algorithms, and digital twins. This holistic approach ensures robust AI systems from end to end.