Skip to content
pitch
NL Nederlands
Menu

AI and Intellectual Property

Advice on the intellectual property questions AI raises, from training data and text and data mining to who owns what a system produces and how you keep your own know-how out of other people's models.

By
By , Attorney-at-Law
Published
Updated

The question

AI raises intellectual property questions at both ends. On the way in: may you, or your supplier, train a model on texts, images, code or data that others own? On the way out: who owns what the system produces, can it be protected, and could it infringe someone else's rights? And in between: what happens to your confidential know-how when your staff paste it into a tool?

For training, EU law has two copyright exceptions for text and data mining, which Directive (EU) 2019/790 defines as any automated analytical technique aimed at analysing text and data in digital form to generate information such as patterns, trends and correlations (Article 2(2)). Article 3 allows research organisations and cultural heritage institutions to mine works they have lawful access to for scientific research. Article 4 allows anyone to make copies of lawfully accessible works for text and data mining and to keep them as long as the mining requires. Under Article 4(3), that general exception does not apply where the rightholder has expressly reserved the use in an appropriate manner, such as by machine-readable means for content made publicly available online. Member States had to transpose the directive by 7 June 2021.

The AI Act (Regulation (EU) 2024/1689) builds on this for providers of general-purpose AI models. They must put in place a policy to comply with Union copyright law, including identifying and respecting reservations of rights under Article 4(3) of the directive, and publish a sufficiently detailed summary of the content used for training, using the AI Office template (Article 53(1)(c) and (d)). Our article on AI, training data and copyright goes into detail.

Output is less settled. Copyright protects creations by people. Whether something produced with an AI tool is protected depends on how much the person behind it shaped the result, and courts are still working out where the line lies. That is why contracts matter more here than in most areas.

Our approach

We start by sorting your position. Are you training or fine-tuning a model, buying a tool that was trained by someone else, using AI to create content, or a rightholder who wants to keep your works out of training sets? Each needs a different answer.

If you train or fine-tune. We look at where the data comes from, which licences and terms apply, whether the text and data mining exception can be relied on and whether rights have been reserved. Databases can carry their own protection (database rights and trade secrets), and licensed datasets have their own terms (data for AI). If you provide a general-purpose model, we help with the copyright policy and the training summary (codes, guidelines and standards).

If you are a rightholder. We help you set up and document a reservation of rights on your website and in your terms, so it is clear and machine-readable, and we advise on what to do when you find your content used.

For output and ownership. Because the law on AI output is unsettled, we settle it by contract: who owns the prompts, the fine-tuning and the results, which licence each party gets, and who bears the risk of third-party claims. When we draft, our Document Studio assembles the terms from our clause library. When a supplier sends its terms, our contract review checks every IP clause against our playbook. Both run on our own AI, and a lawyer reviews every finding and every clause before you see it. Where the deal is wider than IP, see AI contracts.

For your own know-how. A trade secret must have been subject to reasonable steps to keep it secret (Article 2(1) of Directive (EU) 2016/943). Letting staff put confidential material into a tool whose terms allow the supplier to reuse it can weaken that position. We write the rules for staff and check the supplier's terms. Software and code raise their own points (protecting software), and for protecting creative work in general see copyright protection.

What you get

  • A note on your position: which rights are involved, which exceptions apply and where the risks are.
  • For training data: a review of sources, licences and reservations of rights.
  • For rightholders: wording for a reservation of rights on your website and in your terms.
  • Contract clauses on ownership and licences of models, prompts and output, in English, Dutch, French or German.
  • Rules for staff on confidential information and AI tools, which fit into your AI policy.

Limits apply. We cannot guarantee that AI output is protected by copyright, because courts have not settled that. And a reservation of rights tells others what they may not do; it cannot stop every party from ignoring it.

Next step

Tell us how you use AI, which data or content goes in and what comes out, and send any supplier or licence terms you have.

Schedule a meeting or contact us.

In short

  • Training data checked against copyright and database rights
  • Your opt-out from text and data mining set up and documented
  • Ownership of models, prompts and output settled by contract
  • Trade secrets protected when staff use AI tools
  • Reviewed by a lawyer

Our proprietary Legal Tech

Background reading

Let’s talk

Tell us what you’re working on. We’ll get the right person on it and reply within one business day. The first chat is free.

Schedule a meeting