AI Model Auditing Lab
An audit dashboard for overall and per-group performance, error rates, fairness metrics and data quality, auditing your own predictions in your browser, and an eight-area checklist.
العربيةModule experiments (3)
Concepts and formulas
Model auditing
A systematic examination of an AI system: overall and per-group performance, error-rate gaps, fairness metrics, data quality and representation, documentation and oversight (Raji et al., 2020). An audit documents limits; it does not certify.
Model card and datasheet
A model card (Mitchell et al., 2019) states purpose, evaluation data, per-group performance and limits; a datasheet (Gebru et al.) documents data source, collection, composition and suitable uses.
Questions and guides
My model is 95% accurate; is it fair?
Overall accuracy does not say. The average can hide poor performance on a small group, or be high because positives are rare. Compute per-group metrics, error rates and fairness metrics.
Are the lab results a certified audit?
No. Experiments are educational simulations on synthetic data or on your file in your browser; they are not a certified audit nor a final ethical or legal verdict.
Is my data or file sent to the server?
No. Experiments run in your browser and the audit CSV is read locally. "Download PDF" sends only the displayed results (numbers, charts) to this site's server to build the file, which is not stored. Conversations are not stored by default.
Module sources
- Artificial Intelligence Risk Management Framework (AI RMF 1.0)NIST (2023) — Framework
- Fairlearn — open-source fairness toolkitFairlearn contributors (2025) — Tool
- Fairlearn User Guide — AssessmentFairlearn contributors (2025) — Tool
- Model Cards for Model Reporting (arXiv:1810.03993)M. Mitchell et al. (2019) — Paper
- Datasheets for Datasets (arXiv:1803.09010)T. Gebru et al. (2018) — Paper
- Closing the AI Accountability Gap: internal algorithmic auditing (arXiv:2001.00973)I. D. Raji et al. (2020) — Paper
