Model Card
A model card is a structured document describing an AI model's purpose, training data, performance, limitations, and ethical considerations.
What is Model Card?
Model cards were proposed by Mitchell et al. in 2018 and have become a de facto standard for transparency. A model card typically lists the model's intended use, out-of-scope use, training data summary, evaluation results across relevant slices, known biases, and contact information. Hugging Face, Google, and most major model providers publish model cards alongside releases. The EU AI Act requires similar documentation for general-purpose AI models.
How does Model Card apply to enterprise AI?
Enterprises building on top of foundation models should keep their own model cards for fine-tuned models and downstream systems, and they should require model cards from vendors before integration.
Related terms
- Data Card - A data card is a structured document describing a dataset used to train or evaluate an AI model: its source, composition, collection process, intended use, and limitations.
- Transparency Notice - A transparency notice is a clear disclosure to users that they are interacting with an AI system, what it is doing with their data, and what its limits are.
- Explainable AI (XAI) - Explainable AI (XAI) is the set of techniques that make an AI system's outputs and behaviour understandable to humans, supporting trust, debugging, and regulatory compliance.
- EU AI Act - The EU AI Act (Regulation (EU) 2024/1689) is the European Union's horizontal regulation for AI, classifying systems by risk and imposing obligations on providers, deployers, importers, and distributors.
External references
Need help applying Model Card to your enterprise? Submit a short brief and we reply within one business day.