← Back to research
RLHFBy nxted Research Team· Published 29 May 2026· Updated 30 May 2026· 2 min read
EU AI Act readiness: what to demand from your data vendor
If you build a high-risk AI system, your training-data supplier is part of your compliance story. Here is the checklist.
The EU AI Act (Regulation 2024/1689) places real obligations on providers of high-risk AI systems. Your data vendor cannot make you compliant, but the wrong vendor can make compliance much harder.
What to require
- Article 10 (data governance): documented provenance, consent records, and quality controls for every batch of training or evaluation data.
- Annex IV (technical documentation): evaluation methodology, evaluator credentials, inter-rater agreement, and an error taxonomy, in a form you can drop into your technical file.
- Article 14 (human oversight): evidence that real domain experts, not just automated scoring, were the source of ground truth.
- A signed Data Processing Agreement and a clear position on international transfers (UK IDTA, EU SCCs).
What to avoid
- Black-box evaluation where you never learn who reviewed your model or how qualified they were.
- US-only data residency when your training data contains EU personal data.
- Vendors that cannot articulate their position on prohibited practices under Article 5.
Nxted publishes an EU AI Act position statement and ships every Expert report with the credentials and metrics Annex IV expects.
Related reading
What Is RLHF and How Human Evaluation Improves AI Models
RLHF aligns AI models using human judgements. This explainer covers how it works, where it helps, and why who does the evaluation matters.
RLHF Data Providers Compared: Choosing Human Evaluation for Your AI
A neutral guide to the kinds of RLHF and human-evaluation providers, what separates generalist crowds from expert review, and how to choose.
Red-teaming domain AI: why generalist crowds miss expert failures
The most dangerous AI failures are the ones only a domain expert can spot. A generalist crowd will rate them as fine.
n
nxted Research Team
Physical-AI data specialists at OFORO LTD (UK). We write about egocentric data, robotics dataset formats, RLHF and data governance. See what we build.