"Model Cards for Model Reporting" in 2024: Reclassifying Category of Ethical Considerations in Terms of Trustworthiness and Risk Management

Model Cards need updating to reflect "trustworthiness" as a distinct category: The original 2019 Model Cards framework focused on "ethical considerations," but this paper argues that trustworthiness—a concept that has matured significantly in recent years—deserves its own dedicated section. This reflects how the responsible AI field has evolved to define specific, measurable characteristics like accountability, explainability, fairness, and security.

Major global organizations now agree on core trustworthiness principles: The European Commission, OECD, and NIST have independently converged on nine key characteristics that make AI systems trustworthy: accountability, explainability, fairness, privacy, reliability, robustness, safety, security, and transparency. This consensus suggests there is now a shared international understanding of what responsible AI should look like.

One-size-fits-all trustworthiness doesn't work: While these nine characteristics represent a global consensus, the paper emphasizes that "implementation of trustworthy AI varies by context." This means a healthcare AI system and a content recommendation system will need to prioritize and implement trustworthiness differently based on their specific risks and stakeholders.

Better documentation tools help hold AI developers accountable: By reclassifying and expanding how Model Cards document trustworthiness, organizations can create clearer records of how their AI systems were built and what risks they pose. This transparency makes it easier for stakeholders—regulators, users, and affected communities—to understand and evaluate whether an AI system deserves their trust.