Guideline for Trustworthy Artificial Intelligence -- AI Assessment Catalog

AI risks are context-dependent: The paper emphasizes that trustworthiness requirements for AI cannot be applied uniformly across all applications—instead, "specific quality criteria for an AI application depend heavily on the application context," meaning assessment procedures must be tailored to individual use cases rather than following a one-size-fits-all approach.

Machine learning creates new accountability challenges: Because AI systems learn behavior from data rather than following predetermined rules, they pose distinct risks like "unfair treatment of individuals when processing personal data" in high-stakes decisions such as credit lending and hiring—risks that traditional software doesn't typically present.

Abstract principles need practical implementation: The paper identifies a critical gap between high-level trustworthiness concepts discussed by policymakers and business leaders versus concrete, actionable assessment procedures needed to verify whether AI applications actually meet trustworthiness standards in practice.

Assessment must match the technology used: The paper notes that "measures to fulfill [trustworthiness requirements] depend heavily on the AI technology used," suggesting that evaluation methods for a deep learning system would differ significantly from those for rule-based AI systems.

Trustworthiness is essential for AI's potential: The authors argue that AI can only "reach its full potential" when applications are "developed according to high quality standards and are effectively protected against new AI risks," positioning responsible development as necessary for both ethical and economic reasons.