Licence AI and text-and-data-mining clause checker for libraries

Paste an e-resource licence, an amendment or a publisher's terms of use. The checker finds wording on AI use, text and data mining (TDM), copyright exceptions, user data, suspension of access, governing law and confidentiality, highlights it, and suggests what libraries usually ask for. Everything runs in your browser: the text is not sent or stored.

Published 9 October 2026 · Sources checked 9 October 2026

Also available in: Svenska · Nederlands · Norsk · Suomi · Dansk · Français · Español · Português

What is a text and data mining clause?

A text and data mining (TDM) clause is the part of a licence that says whether and how users may analyse the licensed content with software: which methods, through which interface, at what volume, and what they may do with the results. Modern licences often add a separate AI clause covering AI tools and model training.

EU law defines TDM broadly. Article 2(2) of Directive (EU) 2019/790 (the DSM Directive) calls it "any automated analytical technique aimed at analysing text and data in digital form in order to generate information". That covers classic corpus analysis, and it is also why many libraries treat machine learning on licensed content as a form of TDM. Cambridge University Libraries, for example, applies its TDM guidance to generative AI tools. For a fuller explanation of the directive, see our planned page on the DSM Copyright Directive.

Can a licence override the TDM exception?

For scientific research in the EU, no. Article 3 of the DSM Directive gives research organisations and cultural heritage institutions an exception to mine works they have lawful access to, and Article 7(1) states: "Any contractual provision contrary to the exceptions provided for in Articles 3, 5 and 6 shall be unenforceable." A licence can still regulate how mining happens, within limits.

Three details matter when you read a licence:

• Security measures are allowed. Article 3(3) lets rightholders apply measures to protect the security and integrity of their networks and databases, provided the measures do not go beyond what is necessary. Rate limits and an API route can fall under this. The University of Groningen library explains it the same way: publishers may not contractually exclude TDM for research, but may set rules on speed, volume, storage and distribution, and users must follow them (Rijksuniversiteit Groningen, Tekst- en datamining).
• Copies may be kept. Article 3(2) allows copies made for research TDM to be stored securely and retained for scientific research, including to verify results.
• Article 4 is weaker. The general TDM exception in Article 4 applies only if the rightholder has not expressly reserved it (Article 4(3)), and Article 7(1) does not list Article 4. Outside research organisations' scientific research, the licence terms therefore carry more weight.

Outside the EU the picture differs. Jisc says that in the UK, contract terms do not override the non-commercial research TDM exception in section 29A of the Copyright, Designs and Patents Act 1988. In the United States, the UC Berkeley Library warns researchers that "Contract law can override those rights when you agree to specific terms", which is one reason governing law matters.

What AI clauses in library licences say

AI clauses vary widely, and library guidance shows the range. They fall into four broad types.

1. Bans on putting licensed content into AI tools. Stockholm University Library's database terms say: "It is also not permitted to paste text or upload documents containing copyrighted material into AI applications or other automated systems for processing." UCC Library states that use of licensed electronic resources with generative AI systems "including ChatGPT and similar services, is not permitted". The University of Helsinki library lists using protected publications as input to AI tools among the uses that are not allowed.
2. Conditional permission based on where the AI tool runs. The University of Turku summarises the FinELib consortium's Elsevier agreement (2024 to 2025) as requiring the AI tool to run in the subscriber's own environment, or in a third-party environment where uploaded material is only for the subscriber's own users. The University of Fribourg describes Elsevier and Wiley terms that allow closed AI tools that do not train on the data, or self-hosted tools in a secure environment.
3. Training bans with a carve-out. Cambridge University Libraries accepts terms that "prohibit the use of TDM to develop, train, program, improve, and/or enrich any third party Generative AI system" but that do "not prohibit the training of any internal model". Duke University Libraries reports that some licences allow training only in a closed environment where data is not shared with third parties, for non-commercial purposes.
4. Negotiated, detailed permissions. Leiden University Libraries notes that the Dutch 2025 Elsevier agreement contains an extensive TDM and AI clause, with permitted uses and conditions set out in its article 5.4. The agreement text is published on openaccess.nl.

Duke's own answer to whether licensed content can be used for AI training is a fair summary of the situation: "It depends on the individual license agreement."

How to read a licence for AI and TDM terms

1. Collect every document. The agreement in chief, schedules, amendments and the platform's online terms of use. AI wording often sits in the terms of use, which can change without a new signature.
2. Find the definitions. Check how "Authorised Users", "Licensed Materials", "Text and Data Mining" and any AI term are defined. A TDM definition that names machine learning gives users more room than one limited to keyword searching.
3. Look for a savings clause. A sentence such as "nothing in this agreement shall limit statutory exceptions" protects uses the law already allows.
4. Read the TDM clause and the general restrictions together. A generous TDM clause can be undercut by a general ban on robots, crawlers or "systematic downloading".
5. Read the AI clause for scope. Does it restrict training, uploading, outputs, or all AI use? Does it distinguish internal or closed tools from public services?
6. Check liability. Does the library become liable for individual users' breaches? ICOLC's statement says AI clauses should not introduce new liability beyond what was previously agreed.
7. Check suspension and monitoring. Can the publisher cut off the whole institution without notice because of one script?
8. Check data, law and confidentiality. What user data does the platform collect, which law governs, and can you publish the terms?

Reviewing a vendor right now? Validemic checks the vendor's documents against GDPR and the EU AI Act and cites every finding. Try the demo workspace

Negotiation positions libraries publish

Several library bodies publish positions you can cite in negotiations. The table summarises them; follow the links for the full wording.

LIBER described the trend in licence language as wording that "seeks to restrict and prevent AI uses which seems to contradict and undermine EU law". The planned pillar guide on library licences, AI and TDM covers these positions in more depth.

Validemic's analysis
The checker's suggested positions draw on the sources above. Where it offers wording, that is Validemic's example wording to adapt with your legal team, not model clauses endorsed by any of these bodies and not legal advice. A short example for an AI clause: "Authorised Users may use the Licensed Materials with AI tools for non-commercial research, teaching and study, provided the tool does not retain the Licensed Materials or use them to train models made available to third parties."

Privacy terms in content licences

Content platforms process staff and student personal data: login attributes, IP addresses, search and reading histories, cookies and analytics. Licences often say little about this, or point to a general privacy policy. The checker flags data-related wording so you can ask who is controller and who is processor, whether a data processing agreement under Article 28 GDPR applies, and whether usage data is used for the vendor's own product development. Our DPA checker helps with the agreement itself, and the planned guide on library patron privacy and GDPR goes further.

Where the AI Act fits

Article 53(1)(c) of the AI Act (Regulation (EU) 2024/1689) requires providers of general-purpose AI models to "put in place a policy to comply with Union law on copyright and related rights", including identifying and complying with rights reservations made under Article 4(3) of the DSM Directive. This applies to model providers, not to a library signing a content licence. The 2026 amendments in Regulation (EU) 2026/1744 did not change Article 53(1)(c). It does explain why publishers reserve TDM rights and write AI clauses with care. If your institution builds or fine-tunes models on licensed content, read the AI Act guide for universities and the planned AI Act explainer.

What this checker cannot do

• It matches keywords and simple sentence patterns. It does not understand cross-references, definitions or which clause takes priority.
• "Not found" means the checker found no matching wording. It does not mean the licence has no such term: wording can be unusual, in a schedule, or in the online terms of use.
• The labels "restricts", "conditional" and "permits" come from simple word cues such as "may", "not" and "provided that". Treat them as pointers, not conclusions.
• Non-English keyword lists are shorter than the English list. Native-speaker corrections are welcome.

Sources

All sources retrieved 9 October 2026. Short quotations are verbatim from the linked pages.

1. Directive (EU) 2019/790 on copyright and related rights in the Digital Single Market, Articles 2 to 4 and 7, EUR-Lex (read via the Publications Office): eur-lex.europa.eu/eli/dir/2019/790/oj/eng
2. Regulation (EU) 2024/1689 (AI Act), Article 53(1)(c): eur-lex.europa.eu/eli/reg/2024/1689/oj/eng; amending Regulation (EU) 2026/1744: eur-lex.europa.eu/eli/reg/2026/1744/oj/eng
3. ICOLC, Statement on AI in Licensing (22 March 2024): icolc.net
4. LIBER, LIBER Endorses ICOLC Statement on AI in Licensing (10 March 2025): libereurope.eu
5. Jisc, Guidance on resisting restrictive AI clauses in licences (19 June 2024): nationalcentreforai.jiscinvolve.org
6. Samberg, Vollmer and Teremi (UC Berkeley Library), Licensing research content via agreements that authorize uses of artificial intelligence, Authors Alliance (10 January 2024): authorsalliance.org
7. University of Cambridge, Cambridge University Libraries, Text and data mining guidance: libguides.cam.ac.uk/e-resources/tdmguidance
8. Leiden University Libraries, Changes to Read and Publish agreements with publishers as of January 2025 (18 February 2025): library.universiteitleiden.nl
9. Openaccess.nl, Elsevier (2025-2027) publisher deal page with the agreement text: openaccess.nl
10. University of Groningen Library, Tekst- en datamining: rug.nl/library
11. Stockholm University Library, Databases A-Z, terms of use: sub.su.se/en/databases/list
12. Aarhus University Library, Terms of use for using the library's e-resources (June 2023, lists TDM among the per-resource terms): library.au.dk (PDF)
13. University of Helsinki Library, Villkor för användning av e-resurser: helsinki.fi
14. University of Turku, Electronic resources and use of artificial intelligence tools (23 April 2024): utu.fi
15. University College Cork Library, Electronic resources: libguides.ucc.ie
16. UC Berkeley Library, Before you scrape and before you train (2 September 2025): update.lib.berkeley.edu
17. Duke University Libraries, Can licensed e-resources be used for AI training? (updated 11 December 2025): duke.libanswers.com
18. University of Fribourg Libraries, Restrictions on the use of electronic resources with artificial intelligence tools: unifr.ch/biblio

About this page

We built the checker and wrote this page on 9 October 2026, after reading every source listed above on that date. The keyword lists were written by Validemic from the language of the DSM Directive in each language version and from typical licence wording; the invented example licence does not reproduce any real agreement. Library positions are summarised from the bodies' own pages, with short quotations only. The checker is a reading aid for librarians, negotiators and university lawyers. It is not legal advice and does not replace reading the full agreement. If a link has moved, a keyword is missing in your language, or you spot an error, please contact us and we will correct it.