The EU AI Act and your training data supply chain

· 2 min read · compliance, procurement
The EU AI Act places obligations on providers of AI systems that reach backwards into how training data was obtained and documented. In practice that means provenance, consent scope and transparency about automated processing become supplier requirements rather than internal preferences.
The short version
- Obligations on AI system providers reach into training data documentation.
- Article 4 addresses AI literacy for people operating these systems.
- Transparency obligations require disclosure when a person interacts with an AI system.
- The Platform Work Directive affects how data workers themselves are engaged.
- Requirements continue to be clarified, so contracts should anticipate change.
Scoping something like this?
We would rather talk about the problem than quote against a volume figure.
See capabilitiesWhy does this reach your suppliers at all?
Because documentation obligations on a system provider cannot be satisfied without information the data supplier holds. If you must describe how training data was obtained and what consent covered it, you need that from whoever collected it.
Suppliers who cannot produce per-item provenance therefore become a compliance problem rather than merely a quality one.
What should you require contractually now?
Delivery of provenance artefacts with the data rather than retention by the supplier. Explicit consent scope covering commercial training rather than research use. Disclosure of where automated processing is used in the supplier's own operations, including in how they assess and manage their workers.
That last one is increasingly relevant. A supplier making automated decisions about its own contributors has obligations of its own, and those obligations are becoming visible in diligence questionnaires.
- Provenance artefacts delivered, not retained
- Consent scope explicitly covering commercial training
- Disclosure of automated decision-making in supplier operations
- A route to honour data subject requests
- Contract terms that anticipate regulatory change
Where is the honest uncertainty?
Plenty of it, and a supplier claiming complete certainty about their compliance position is overstating. Guidance continues to develop and reasonable interpretations differ on specifics.
What is not uncertain is the direction: documentation, provenance and transparency requirements are increasing rather than decreasing. Building for them now is cheaper than retrofitting.
What about the workers themselves?
The Platform Work Directive addresses how platform workers are engaged and how algorithmic management is applied to them, which is directly relevant to how a data supplier operates.
A supplier that makes automated accept and reject decisions about contributors, with no human in the loop and no stated reason, is carrying a risk that becomes yours through the supply chain.
People also ask
Does the EU AI Act apply to training data suppliers?
Obligations fall on providers of AI systems, but satisfying them requires documentation only the data supplier holds. In practice provenance and consent documentation become supplier requirements.
What should I add to supplier contracts because of the AI Act?
Delivery of provenance artefacts with the data, explicit consent scope covering commercial training and disclosure of where the supplier uses automated decision-making in its own operations.
Does the AI Act affect how data workers are managed?
The Platform Work Directive addresses algorithmic management of platform workers directly, which is relevant to any supplier making automated decisions about its own contributors.