Guides
Straight answers on legal data for AI.
Each guide answers one question a buyer or ML engineer asks, cites the primary sources and says where the law is unsettled. None of it is legal advice.
Last updated
-
AI Act Article 53 training-data summary: a template for legal corpora
Under Article 53(1)(d) of the AI Act, providers of general-purpose AI models must publish a training-content summary using the AI Office template of 24 July 2025; a licensed legal corpus such as Arieta Data is reported in Section 2.2 (private datasets obtained from third parties), while a model trained only on the free public sample, whose records carry their upstream licences, would report it under Section 2.1 (publicly available datasets).
-
CELEX, ECLI and ELI explained: legal identifiers for ML engineers
CELEX is EUR-Lex's identifier for every EU legal document (the GDPR is 32016R0679); the European Case Law Identifier (ECLI), introduced by Council conclusions in 2011, identifies court decisions across the EU (Schrems II is ECLI:EU:C:2020:559); and the European Legislation Identifier (ELI), from Council conclusions in 2012, gives legislation stable URIs such as http://data.europa.eu/eli/reg/2016/679/oj. In a pipeline they are the join keys, citation targets and deduplication keys that make legal data verifiable.
-
Citation graphs as supervision signal for legal retrieval
A legal citation graph links documents by explicit, typed relations such as amendment, repeal, legal basis and transposition, so its edges can serve as positive pairs, hard-negative sources and gold relevance sets for legal retrievers without manual labelling; Arieta Data's graph dataset types every edge and records where it came from, with no probabilistic edges.
-
Common Crawl vs curated legal data for LLM training
Common Crawl is free and huge (over 300 billion pages, per its homepage in November 2026), but it holds legal text as page snapshots with menus, no article structure, no identifiers or versions, uneven coverage, and no licence of its own; many official legal portals also block CCBot or AI crawlers. For legal pre-training breadth it helps; for fine-tuning, RAG and evals, a curated corpus with structure, dates and per-source licences is what makes results verifiable.
-
EU copyright, TDM opt-outs and the AI Act: what applies to legal training data
Partly. Article 4 of Directive (EU) 2019/790 allows text and data mining of lawfully accessible works unless the rightholder has reserved it, and Article 53(1)(c) of the AI Act requires general-purpose AI providers to have a policy that honours those reservations; but statutes and judgments are outside copyright in most member states, and EUR-Lex content is licensed for reuse, so for official legal texts the opt-out usually has little to bite on. Editorial content, third-party material and personal data still need care.
-
EU directive transposition as structured data: national measures, CELEX sector 7 and the graph
Member states notify the national measures that transpose each directive to the European Commission, and the Publications Office publishes them on EUR-Lex under the directive's 'National transposition' tab, each with a CELEX sector 7 number such as 72016L0680FRA_263118; harvested from Cellar, these notifications become directive-to-national-law edges. A notification is the member state's own statement, not proof of complete or correct transposition. Arieta Data's 2026-11 graph holds 66,670 transposition edges.
-
EU legislation datasets for LLM training: what's available and what it costs
Free research datasets such as MultiEURLEX, EURLEX57K and EUR-Lex-Sum are static, task-shaped snapshots; building your own from EUR-Lex and Cellar is free in licence terms but costly in engineering; licensed corpora such as Arieta Data add article-level structure, version history, a typed citation graph and monthly updates, from €7,000 a year.
-
How to build a legal eval set from a citation graph
Derive the gold answers from relations the law already states: a citation graph gives cited provisions, amending acts, implementing acts and notified national transposition measures as gold sets, and a version history gives as-of dates and held-out changes; lawyers then review a sample instead of labelling everything. Arieta Data's 2026-11 graph has 801,149 typed edges, including 66,670 transposition edges.
-
How to chunk legal documents for RAG: by article, not by token count
Chunk legislation by its own structure: one chunk per article (or per paragraph for long articles), carrying the act's identifier, title, article number and version date as metadata; fixed token windows split rules from their exceptions and lose the citation a grounded answer needs, while article-level chunks map one-to-one to citable provisions. Arieta Data's 2026-11 snapshot ships approx. 4.07M articles already split this way.
-
Is EUR-Lex data licensable for AI training? The EUR-Lex legal notice and Decision 2011/833/EU explained
Yes, in general: the EUR-Lex legal notice (Commission documents: Decision 2011/833/EU) allows the legal documents on EUR-Lex to be reused, and Commission Decision 2011/833/EU allows Commission documents to be reused for commercial or non-commercial purposes without application, on conditions such as acknowledging the source and not distorting the meaning; it does not mention AI training, and it excludes third-party intellectual property and does not displace data-protection law.
-
National legislation datasets in Europe: sources, licences and formats by country
Every EU member state and the UK publishes its legislation online, and most offer an API or bulk XML: Légifrance/DILA (France, Licence Ouverte 2.0), legislation.gov.uk (UK, Open Government Licence v3.0), gesetze-im-internet.de (Germany, statutes outside copyright under § 5 UrhG), BOE (Spain), RIS (Austria, CC BY 4.0) and others; terms differ by country, so attribution, authenticity and personal-data rules must be tracked per source. Arieta Data's 2026-11 snapshot holds 350k+ national legislation documents from 15 countries.
-
Parallel corpora of EU law in 24 languages for MT and multilingual LLMs
EU legislation is published in all 24 official EU languages, which are equally authentic, so EUR-Lex is the largest source of professionally translated legal text in Europe; ready-made parallel corpora include the Commission's DGT-TM (24 languages, latest release 2021), JRC-Acquis (22 languages) and MultiEURLEX (65k laws in 23 languages), and you can build your own by aligning language versions on the CELEX number and article number.
-
Point-in-time legal data: why version history matters for legal AI evals
Legal texts are amended after an eval is written, so an eval built on the latest text leaks future law into past questions and silently invalidates gold answers; Arieta Data's history dataset records every content-hash version it has observed with a timestamp, so an eval can be pinned to what a source said on a given date.
Questions the guides do not answer?
Ask us about your jurisdictions, licensing or AI Act summary. Or get a sample cut and see the data for yourself.
- We reply within one business day
- Custom cuts in 5–6 business days
- Free sample, no sign-up