📌 1. Overview: AI Data Training Rights & Licensing

AI models require large datasets for training, which often include copyrighted, personal, or proprietary content. Using these datasets raises several legal issues:

Copyright infringement: Using copyrighted works (text, images, music) without authorization may constitute infringement.

Database rights: UK and EU law protect structured collections of data. Using such data without a license may violate database rights.

Data privacy compliance: Personal data in training datasets must comply with UK GDPR.

Licensing obligations: Companies must secure appropriate licenses for datasets, especially commercial use.

Contractual restrictions: Third-party agreements may restrict dataset usage or derivative works.

Failure to address these issues exposes UK companies to legal, financial, and reputational risk.

📌 2. Key Legal Principles

2.1 Copyright in Training Data

Using copyrighted works for AI training may be infringing if no license or exemption applies.

UK copyright law recognizes exceptions (e.g., research or text/data mining for non-commercial purposes) but commercial use typically requires permission.

2.2 Database Rights

UK law protects databases that demonstrate substantial investment in obtaining, verifying, or presenting data.

Extracting or reusing substantial parts of a database without a license may constitute infringement.

2.3 Personal Data Compliance

AI models processing personal data must comply with UK GDPR:

Lawful basis for processing

Data minimization

Transparency and accountability

2.4 Licensing Obligations

Companies must document terms of use for each dataset.

Open-source or Creative Commons datasets often impose restrictions on commercial use or derivative works.

2.5 Risk Mitigation

Maintain audit trails of dataset provenance.

Conduct legal reviews of dataset licenses before commercial deployment.

Use synthetic or anonymized data where possible.

📌 3. Relevant Case Law & Legal Precedents

Below are six UK and international cases relevant to AI data training rights and licensing:

1) Getty Images v. Stability AI (2025)

AI trained on Getty Images’ copyrighted works without license.

Court recognized potential copyright infringement despite AI-generated outputs being “new.”

Implication: UK companies must secure licenses for copyrighted material used in training AI.

2) SAS Institute Inc v. World Programming Ltd (2013, UKSC)

Reproducing functionality of software without copying source code was allowed, but copying underlying datasets can still infringe.

Implication: Functional imitation does not eliminate copyright or licensing obligations for data.

3) British Horseracing Authority v. William Hill (2020)

Database right infringement for extracting betting data without permission.

Implication: AI training on commercial datasets may require explicit licensing.

4) Clearview AI Enforcement (ICO, 2025)

Scraping images from public websites violated data protection law.

Implication: Personal data in training datasets triggers GDPR obligations; consent or lawful basis is required.

5) Nova Productions Ltd v. Mazooma Games Ltd (2006)

Copying substantial parts of digital works constituted copyright infringement.

Implication: AI training on copyrighted content without a license is risky even if outputs differ.

6) Thaler / DABUS Case (UKSC, 2023)

AI cannot hold IP rights; liability and ownership fall on human operators.

Implication: Companies using AI trained on third-party data remain responsible for licensing compliance and copyright clearance.

📌 4. Practical Steps for UK Companies

Dataset Licensing Review

Verify ownership and terms of use before training AI.

Secure explicit licenses for commercial use.

Copyright Compliance

Avoid unlicensed copyrighted material.

Consider fair use/text and data mining exceptions carefully.

Database Rights Compliance

Review substantial parts of structured datasets for licensing obligations.

Personal Data Governance

Conduct Data Protection Impact Assessments (DPIAs).

Ensure anonymization or pseudonymization where possible.

Audit and Documentation

Maintain provenance logs for all datasets.

Document AI model training workflows and licenses.

Synthetic and Open Datasets

Use synthetic or fully licensed open datasets to reduce risk.

📌 5. Summary Table: AI Training Data Rights & Licensing