Risk · R-146 · Intellectual Property & Legal Liability
Copyright infringement in training data
A model is trained or fine-tuned on protected works without a licence or an applicable exception, by the organisation or by a provider whose model it uses.
- Severity
- High
- Audience
- Board · Heads-of · Practitioner
- Frameworks
- Article 53 · Intellectual Property
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-11
What it is
Training a model involves copying the material it learns from. Where that material is protected by copyright, copying needs the owner's permission or a legal exception. Whether and when exceptions apply to AI training is contested and varies by country: some jurisdictions allow text and data mining subject to an opt-out by rights holders, others allow it only for limited purposes, and litigation is under way in several. An organisation is exposed directly when it trains or fine-tunes on material it has no right to use, and indirectly when it relies on a provider's model whose training is challenged.
How it happens
- Data is scraped from the web on the assumption that public means free to use.
- Licensed content is used for training although the licence covers reading only.
- Rights holders' opt-out signals are ignored or not checked.
- A provider's model is adopted with no enquiry into how it was trained.
- Internal documents used for fine-tuning include third-party material.
Why it matters
Claims can be large and the remedies severe, including orders to stop using a model. An organisation dependent on a provider's model shares the disruption if that model is withdrawn or restricted. The EU AI Act requires providers of general-purpose models to have a policy for complying with copyright law and to publish a summary of training content.
Warning signs
- No record shows the rights position of a training dataset.
- Web-scraped data is in use with no legal review.
- Provider contracts contain no indemnity for intellectual property claims.
- The provider will not say what its model was trained on.
Who owns it
Legal owns the assessment of rights. The data owner and model owner are accountable for what they train on. Procurement secures warranties and indemnities from providers.
In practice
A publisher fine-tunes a model on its archive to generate summaries. A large part of the archive is agency copy and freelance work licensed for publication only. The same pattern generalises to any organisation that treats the content it holds as content it may train on.
Mitigating controls
Primary controls
- Intellectual property and licence clearance Preventive
- Checks rights in training data before use. Training data, models, components and significant outputs are checked for rights and licence terms before they are used, published or shipped.
Supporting controls
- AI contract terms Directive
- Obtains indemnities. Contracts with AI suppliers cover data use, change notification, information and audit rights, sub-contracting, incident notification, liability and exit.
- Dataset documentation Directive
- Records the rights position of each dataset. Each dataset used for AI has a maintained record of its source, collection method, contents, rights, lawful basis, known limitations and permitted uses.
- Third-party model and vendor due diligence Directive
- Establishes how a provider's model was trained. Before an AI supplier, model or dataset is adopted, it is assessed against AI-specific criteria to a depth set by how material the use is, with evidence for the answers.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| EU AI Act | Article 53 | Obligations for providers of general-purpose AI models Copyright policy for general-purpose models. |
| NIST Generative AI Profile (NIST AI 600-1) | Intellectual Property | Intellectual Property The intellectual property risk. |
Sources
- EU AI Act: Regulation (EU) 2024/1689 European Union (EUR-Lex) · official text
- NIST Generative AI Profile NIST · framework
Related terms
How to cite
Axiom Verity (2026). Copyright infringement in training data. Reference Library, R-146. https://axiom-verity.com/risks/copyright-infringement-in-training-data. Reviewed 2026-11-11.
Found an error or a better source? Report a correction.