INTRA/BA/BA009Model validation and bias testing
The examination of an assessment method for accuracy and for differential effect before and during its use.
| Control | Jurisdiction | Source | What it establishes | Prerequisite or recommendation |
|---|---|---|---|---|
| BA009/EUMandatory | European Union | Regulation (EU) 2024/1689, articles 5(1)(f), 10, and 26(7), and Annex III(4)European Parliament and Council · read 11 Aug 2026 | That inferring emotions at work is prohibited, that employment and worker management are a high-risk category, what the data a high-risk system is trained and tested on have to answer for, and that representatives are told before deployment. | The training, validation, and testing sets behind a high-risk system fall under data governance practices, and what those have to cover is listed: the design choices, where the data came from and, for personal data, what they were originally collected for, the assumptions about what the data are supposed to measure, an examination for biases likely to bear on health and safety, to affect fundamental rights adversely, or to lead to prohibited discrimination, especially where the outputs feed the inputs of later operations, and measures to detect, prevent, and mitigate what the examination finds. The sets are to be relevant, sufficiently representative, and so far as possible free of errors and complete, with statistical properties appropriate to the people the system is to be used on. |
| Reported | United Kingdomwhere written | Integrating Human Factors into Insider Threat Detection: A Systematic ReviewPathirana, Roberts, Kalutarage, and McDermott · read 16 Aug 2026 | The models, data sources, and evaluation methods used where detection research has taken up human factors. | The evaluation methods used across the published work have been mapped, which is the starting point for validating one. |