AI lineage (data and model provenance)

Why it matters

• The EU AI Act requires data governance for high-risk systems covering data collection processes and the origin of data (Article 10), and providers of general-purpose AI models must publish a summary of the content used for training (Article 53).
• Without lineage, a poisoned dataset, a licence problem or a biased source cannot be traced to the systems it affected.
• It makes model changes auditable, which is needed to decide whether a change is a substantial modification.

In detail

Lineage covers datasets (origin, consent and licence, cleaning and labelling), models (base model, fine-tuning, weight versions) and the pipeline between them. It is the evidence behind data governance, copyright and privacy compliance, incident investigation and reproducibility. Content provenance, marking AI-generated outputs as such, is a related but separate practice.

Laws and policies on record that use it

• Singapore Model AI Governance FrameworkSingapore · Voluntary standard
• PDPC AI advisory guidelinesSingapore · Guidance

Duties that mention it

• Manage data quality, model development and monitoring across the lifecycleSingapore Model AI Governance Framework · Singapore

Related concepts

• Data poisoningData poisoning is an attack that corrupts an AI model by inserting or altering its training or fine-tuning dat...
• AI bill of materials (AIBOM)An AI bill of materials (AIBOM) is a machine-readable inventory of the components of an AI system: models, dat...
• Substantial modificationA change to an AI system after it is placed on the market that the provider's original conformity assessment d...
• Red teaming (AI)AI red teaming is structured adversarial testing in which testers try to make an AI system fail, by producing...