Skip to content
←Back to glossaryAI

Data governance

The set of practices governing the quality, representativeness and traceability of the data used to train and operate an AI system.

Definition

Regulation (EU) 2024/1689, Article 10, published in the Official Journal on July 12, 2024: the training, validation and testing data sets of a high-risk AI system must be relevant, sufficiently representative and, to the best extent possible, free of errors and complete in view of the system's intended purpose; their origin, annotation, cleaning and the examination of possible biases are documented. Data governance covers the rules and procedures an organization uses to control its data from collection to use. It looks after their quality, representativeness, relevance and traceability. For high-risk AI systems, the AI Act makes it an explicit obligation, because a model is never more reliable than the data it was trained on. Biased, incomplete or unrepresentative data produce algorithmic bias and, in insurance, a risk of discriminatory pricing. Rigorous governance catches and corrects these flaws upstream, documents datasets and checks their compliance, notably with the GDPR. It is one of the pillars of regulatory compliance, which has become a condition of insurance underwriting.

Its 9 cross-references, woven across the glossary's 18 categories
Data governance: AI glossary illustration, AlgoPolis
Example

Textbook case: before deploying a health pricing model, an insurer audits its training data to verify that no proxy variable introduces indirect discrimination, a data-governance requirement set by the AI Act.

Read next
AIAlgorithmic biasA systematic error of an AI system that produces results unfairly unfavorable to certain groups, often arising from biased training data.Read the entry →AINIST AI Risk Management FrameworkA voluntary American framework organizing AI system risk management into four functions, now the common vocabulary of audits.Read the entry →AIISO/IEC 42001A management system standard for artificial intelligence that a third party can certify. It covers the organization and says nothing about the quality of any model.Read the entry →
Sources
  • Regulation (EU) 2024/1689 (AI Act), Article 10, Official Journal of the European Union, 12 July 2024
Related terms
Related articles
Stay informed

Follow the research

One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.

No spam, one-click unsubscribe, no data ever sold.