Is EUR-Lex data licensable for AI training? The EUR-Lex legal notice and Decision 2011/833/EU explained

Yes, as a general rule. The EUR-Lex legal notice allows the legal documents published on EUR-Lex to be reused for commercial or non-commercial purposes, with acknowledgement of the source, and states that the Commission’s reuse policy is based on Decision 2011/833/EU. For Commission documents, Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents provides that “all documents shall be available for reuse”, held by the Commission or by the Publications Office on its behalf, “for commercial or non-commercial purposes”, in principle without charge and without the need for an individual application (Article 4). The conditions are modest: acknowledge the source, do not distort the original meaning, and accept that the Commission is not liable (Article 6(2)). The Decision does not mention machine learning, so whether AI training is covered is a question of interpreting “reuse”, which the Decision defines broadly. The main limits are third-party intellectual property, documents of other institutions (Parliament and Council legislation is reused under the EUR-Lex notice rather than the Decision; CJEU case law under the CURIA notice), and personal data.

What does Decision 2011/833/EU actually say?

Decision 2011/833/EU sets the Commission’s open reuse policy and replaced the earlier Decision 2006/291/EC. It was published in the Official Journal on 14 December 2011 (OJ L 330, p. 39). The provisions that matter for a training-data pipeline are these:

Article	Rule	Relevance for AI training
Art. 2(1)	Applies to public documents produced by the Commission or on its behalf, published or not	Covers documents the Commission (or the Publications Office on its behalf) produces and publishes, including Commission acts and proposals. Legislation adopted by Parliament and Council, and CJEU case law, are reused under the EUR-Lex legal notice and the CURIA notice respectively
Art. 2(2)	Does not apply to software, industrial property (trade marks, logos), or documents the Commission cannot license because of third-party IP	Standards referenced in acts, images and similar third-party material are outside the grant
Art. 2(4)	No authorisation to reuse documents “in a manner calculated to deceive or to defraud”	Do not present a model’s output as the official text
Art. 3(2)	“Reuse” means use “for commercial or non-commercial purposes other than the initial purpose for which the documents were produced”	Broad enough on its face to include building training, retrieval and eval datasets
Art. 4	Reuse without application and without charge. The Decision is implemented in full respect of data-protection rules	No permission request needed; your own GDPR obligations apply in full
Art. 6(2)	Conditions may include acknowledging the source, not distorting the meaning, and non-liability of the Commission	The practical obligations you pass through to a dataset and model card
Art. 9	Reuse is “in principle free of charge”	Licence fees are not the cost of EUR-Lex data

In February 2019 the Commission also adopted CC BY 4.0 as the default licence for its content and CC0 for raw data and metadata. The EUR-Lex legal notice, as quoted in the MultiEURLEX dataset card in 2021, says the EUR-Lex editorial content, the summaries of EU legislation and the consolidated texts are licensed under CC BY 4.0, so reusers must acknowledge the source and indicate changes. Check the current wording on the EUR-Lex legal notice before you rely on it.

Does “reuse” include training an AI model?

Decision 2011/833/EU does not mention AI, machine learning or text and data mining, because it predates them as policy topics. Its definition of reuse in Article 3(2) is purpose-neutral: any commercial or non-commercial use other than the purpose the document was produced for. Training a model, indexing for retrieval and building benchmarks all fit that wording on a plain reading, and nothing in the Decision carves them out.

What is not settled is the wider copyright question for AI training in the EU. The Court of Justice has not ruled on how the text and data mining exceptions in Directive (EU) 2019/790 apply to model training. Advocate General Szpunar’s opinion of 3 November 2026 in Case C-250/25 Like Company is not binding, and the judgment may not decide that point at all. For EUR-Lex content this matters less than for news or books, because the rightholder has granted an express reuse licence: the question is the scope of that licence, not whether an exception applies. How the TDM opt-out and the AI Act copyright policy apply to official texts generally is covered in EU copyright, TDM opt-outs and the AI Act for legal data.

What conditions apply in practice?

Three conditions from Article 6(2) and the CC BY 4.0 notice translate into concrete pipeline obligations.

Acknowledge the source. Keep a source and attribution field on every record, and carry it into your dataset documentation and, where relevant, your AI Act training-content summary. A line such as “Source: EUR-Lex, © European Union” alongside the reference to Decision 2011/833/EU is the common form.

Do not distort the meaning. Parsing errors that merge or truncate articles, or text presented as current when it has been amended, can distort meaning. Version history and clear labelling of consolidated texts as non-authentic reduce that risk. Only the Official Journal (and, for case law, the Reports of Cases) is authentic.

Indicate changes. Under CC BY 4.0 and the CURIA notice, say what you changed: text extracted from HTML, normalised whitespace, split into articles.

What about CJEU case law?

CJEU judgments are published on EUR-Lex and on curia.europa.eu. The CURIA copyright notice states that the judicial decisions in its case-law database may be reproduced. For editorial content the notice asks for credit to the “Court of Justice of the European Union”, no distortion of meaning and an indication of changes; Arieta Data applies the same conditions to case law. CJEU case law is in Arieta Data’s licensed corpus, though not in the free open sample; a CJEU case-law excerpt is available under an evaluation licence on request. The notice also warns that judgments sometimes contain third-party protected works, such as images of trade marks, which need the rightholder’s permission. Judgments name parties and other individuals, so data-protection law applies to how you process them.

What does the Decision not cover?

Decision 2011/833/EU is a Commission instrument. Four gaps matter for a training corpus:

1. Other institutions’ documents. Recital 14 says the Decision should not apply to documents received from other institutions. EUR-Lex also hosts Parliament and Council documents, so check the EUR-Lex notice for those.
2. Third-party IP. Material such as referenced standards or embedded images is excluded by Article 2(2)(b).
3. Personal data. Commission decisions in competition and state-aid cases, and court judgments, can identify individuals. The Decision is implemented in full respect of data-protection rules, but that is an instruction to the Commission: reusers are bound by the GDPR or UK GDPR directly, and the EDPB Opinion 28/2024 on AI models sets out how supervisory authorities assess AI training on personal data.
4. National law. Member-state legislation and case law are governed by national rules, not by the Decision.

How do national sources compare?

National legal sources are licensed source by source, and the terms differ more than EUR-Lex’s do. A few examples from the sources Arieta Data documents:

Source	Terms	Point to note
EUR-Lex (EU)	EUR-Lex legal notice (Commission reuse policy, Decision 2011/833/EU); CC BY 4.0 for editorial content and consolidated texts as quoted on the MultiEURLEX card	Attribution, no distortion
Austria, Finland, Slovenia, Lithuania, Italy (Normattiva)	CC BY 4.0 (Normattiva since 1 January 2026, after a CC BY 4.0 NC experimental phase; Arieta Data’s IT records were all retrieved in 2026)	Verbatim text stays CC BY in every recipient’s hands
France (Légifrance)	Licence Ouverte 2.0	Source and date of last update
Spain (BOE)	AEBOE reuse conditions	Derived products must state “Basado en datos de la Agencia Estatal Boletín Oficial del Estado”
UK legislation	Open Government Licence v3.0	Personal data excluded from the licence
Italy, Corte costituzionale	CC BY-SA 3.0	Share-alike
UK case law (Find Case Law)	Open Justice Licence	Computational analysis, including “machine learning or AI training on judgment datasets”, needs a separate licence
French case law	Licence Ouverte 2.0 plus statute	Art. L111-13 Code de l’organisation judiciaire makes profiling judges and court clerks from their identity data a criminal offence

The country-by-country picture is in National legislation datasets in Europe, and the per-lane detail is on the coverage page, including the EU lane, UK case law, Corte costituzionale and France.

Key facts

• Commission Decision 2011/833/EU of 12 December 2011 (OJ L 330, 14.12.2011, p. 39) makes Commission documents available for commercial and non-commercial reuse, in principle free of charge and without individual application.
• The EUR-Lex legal notice applies that policy to the legal documents on EUR-Lex; CJEU case law is reused under the CURIA copyright notice.
• The conditions are to acknowledge the source, not distort the meaning, and indicate changes; the Commission accepts no liability.
• Third-party intellectual property, such as referenced standards and images, is outside the grant, and data-protection law applies in full.
• The Decision does not mention AI training; its definition of reuse is purpose-neutral.

Frequently asked questions

Does Decision 2011/833/EU cover regulations adopted by the Parliament and Council?

Not directly. The Decision covers documents produced by the Commission or on its behalf, and recital 14 excludes documents received from other institutions. Legislation adopted by the Parliament and Council is reused under the EUR-Lex legal notice.

Do I need to ask permission to train on EUR-Lex?

No. Reuse under the Commission’s policy needs no individual application. You must acknowledge the source, not distort the meaning, and comply with data-protection law for any personal data.

What attribution should a dataset or model card use?

A line such as “Source: EUR-Lex, © European Union” with a reference to the EUR-Lex legal notice, plus a note of the changes made (for example: text extracted and split into articles). For CJEU case law, credit the Court of Justice of the European Union.

Is CJEU case law free to reuse for AI training?

The CURIA copyright notice states that the judicial decisions in its case-law database may be reproduced. Some judgments contain third-party protected works, such as trade-mark images, and judgments name people, so data-protection law applies.

How Arieta Data handles the licence chain

Arieta Data’s EU lane is built from EUR-Lex and Cellar (in snapshot 2026-11, 61,410 documents, of which 22,888 are legislation and 24,180 CJEU case law), reused under the EUR-Lex legal notice (and, for CJEU case law, the CURIA notice), and every record carries its source, licence, licence URL and attribution text. Each snapshot ships a LICENCES.md with the attribution text per upstream source. Raw upstream HTML is never shipped. UK case law is not offered, Corte costituzionale decisions (CC BY-SA 3.0) are national case law, delivered masked as a separate CC BY-SA file, French restrictions pass through to licensees, and documents withdrawn upstream arrive as tombstones that licensees must delete. National case law is included, delivered with personal data masked.

The Arieta Data licence agreement is available on request. Request a sample: we reply within one business day, and a custom cut by jurisdiction, language, date range or document type is ready within 5⁠–⁠6 business days. Rights by tier are on the licensing page, and the sourcing method is on the provenance page.

Not legal advice. This guide summarises public sources as of 13 November 2026; take your own legal advice before training on any dataset.