Data governance under Article 10

Article 10 is the bridge between Data Governance and the AI Act. It requires the training, validation and testing datasets of high-risk systems to be managed with appropriate data governance practices.

That includes documenting design choices, the origin of the data and its original purpose; preparation operations (labelling, cleaning, enrichment); assumptions about what the data measures; an assessment of its availability and sufficiency; and an examination of possible biases with measures to detect and correct them. Data must be relevant, sufficiently representative and, as far as possible, free of errors and complete.

That is why we say AI is only as reliable as its data: without owners, quality rules and lineage, this article is very hard to meet. After the Digital Omnibus, the exceptional processing of sensitive data to correct bias moved to a new Article 4a.