AI Glossary Term

What does “Model Card” mean?

Definition

Documentation describing a model's intended use, metrics, limitations, and risks.

Browse the full AI GlossaryExplore all free AI guides

How Model Card works

Diagram for Model Card, layers from top to bottom: What it is and who made it; Intended and out-of-scope uses; Training data and evaluation results; Limitations, risks, and bias findings.What it is and who made itIntended and out-of-scope usesTraining data and evaluation resultsLimitations, risks, and bias findings

Example

A model card for a face-detection model might report accuracy separately for different skin tones and lighting conditions, and say it should not be used for identifying individuals.

Why it matters

A model card gives users a concise record to judge whether a model fits their use. The idea was proposed in a 2019 paper by researchers at Google.

Common questions

Who writes model cards?

Usually the team that built the model, though independent reviewers can add findings.

Are they required?

Mostly voluntary, though documentation requirements in some regulations are making similar disclosures more common.

How much can I trust one?

It is the developer's account, so it is most useful when supported by independent testing.

Related terms

AI SafetyA field focused on reducing harmful behavior, failures, and misuse risks in AI systems.Algorithmic TransparencyHow clearly an AI system's logic, data sources, and limitations are documented and understandable.Feature ExtractionConverting raw data into informative features that a model can use.JailbreakA prompt technique intended to bypass a model's safety constraints.Red TeamingStress-testing an AI system with adversarial prompts to reveal failures and risks.Tool UseA model's ability to call external tools such as search, calculators, or APIs.

Learn more in our free guides

AI Models ExplainedModel CollapseModel LifecycleModel Context Protocol

Test yourself

AI Models Explained QuizModel Collapse QuizModel Lifecycle Quiz

See also

Mixture of Experts (MoE)Model DriftMemory (Agent Memory)Model QuantizationMachine Learning (ML)Multimodal ModelLoss FunctionNamed Entity Recognition (NER)