Version it with the data
Write a new version whenever the data changes: a relabelling pass, a new source, a filter. The change log in section 7 is what an auditor reads first.
Free tool · datasheet generator
EU AI Act dataset documentation generator
Fill in a datasheet for a training, validation or test dataset, structured around Article 10 and Annex IV of the EU AI Act, then download it as Markdown or save it as a PDF.
- Structured around Article 10 and Annex IV
- Drafts the GPAI training summary
- Stays in your browser
A documentation aid, not legal advice. Check obligations against the text of the AI Act as amended, and with your counsel.
01 · What the Act asks
Data governance under Article 10
High-risk AI systems must be developed on training, validation and testing data that is relevant, sufficiently representative and, as far as possible, free of errors and complete for its intended purpose. Article 10(2) lists the practices to document.
| Article | What to document |
|---|---|
| 10(2)(a) | Design choices |
| 10(2)(b) | Data collection processes and the origin of the data; for personal data, the original purpose of collection |
| 10(2)(c) | Preparation: annotation, labelling, cleaning, updating, enrichment, aggregation |
| 10(2)(d) | Assumptions about what the data measures and represents |
| 10(2)(e) | Availability, quantity and suitability of the data |
| 10(2)(f) | Examination of possible biases |
| 10(2)(g) | Measures to detect, prevent and mitigate those biases |
| 10(2)(h) | Data gaps or shortcomings, and how they are addressed |
| 10(3) | Relevance, representativeness, errors, completeness, statistical properties |
| 10(4) | The geographical, contextual, behavioural or functional setting of use |
| Annex IV 2(d) | Datasheets: general description, provenance, scope and main characteristics, how data was obtained and selected, labelling procedures, data cleaning |
Where the deadlines stand
Obligations for providers of general-purpose AI models apply, including the public training-content summary (Article 53).
Deadline for general-purpose AI models placed on the market before 2 August 2025.
High-risk obligations, Article 10 included, apply to Annex III systems.
High-risk obligations apply to AI in Annex I regulated products.
High-risk dates as amended by the Digital Omnibus on AI, Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026. Sources: Regulation (EU) 2026/1744, AI Act Explorer: Digital Omnibus, AI Office template.
02 · Keep it current
A datasheet is only useful if it stays true
Link it to the model
Dataset Finder’s training-data inventory (Pro) maps each project to its datasets, with lineage, licences, governance metadata and an audit trail.
Record who labelled what
Custom data ordered through Dataset Finder comes with a record of who annotated what, the guidelines used and the QA results, ready for section 3.
03 · FAQ
Questions about AI Act documentation
Does every dataset need this documentation?
Article 10 applies to the training, validation and testing data of high-risk AI systems, and Annex IV asks for datasheets in their technical documentation. For other systems it is not mandatory, but the same record answers the questions customers, auditors and GDPR reviews ask.
Is a completed datasheet enough to comply?
No. It documents the data governance that Article 10 requires; it does not replace the practices themselves, the rest of the technical documentation, or a legal review. Treat it as evidence you keep up to date, not a certificate.
What changed with the 2026 Digital Omnibus?
The Omnibus (Regulation (EU) 2026/1744), published in the Official Journal on 24 July 2026, moved the high-risk obligations: systems in Annex III now apply them from 2 December 2027, and AI in Annex I regulated products from 2 August 2028. It amended parts of Article 10, moved the rules on using special-category data to detect and correct bias into a new Article 4a, which now also covers other AI systems and deployers, and lets SMEs and small mid-caps provide the technical documentation in a simplified form. What Annex IV asks you to say about the data is unchanged.
Do general-purpose AI model providers need something different?
Yes. Under Article 53(1)(d) they publish a summary of training content, and using the AI Office’s template of 24 July 2025 is mandatory. It has three parts: general information (the model, and the modalities and size of its training content), the data sources by type, including the top domains of crawled data, and data processing, including how text-and-data-mining opt-outs were respected. Section 8 of the generator helps you draft that content; publish it in the template itself.
Where is my data stored?
In your browser only. The form saves a draft in this browser so you can come back to it, and nothing is sent to Dataset Finder. Download the Markdown or a .json copy to keep it with your project.
04 · Keep reading
Guides on compliant training data
More free tools
Other tools for AI training data
How much data do I need?Pilot, workable and strong ranges for your fine-tuning task, with sources.Annotation cost estimatorCredits, euros and plan discounts, compared with labelling in-house.Annotation guidelines generatorLabelling guidelines with definitions, edge cases and format, exported to Word or PDF.Cohen’s kappa calculatorInter-annotator agreement and weighted kappa, with a confidence interval.