AI copyright licensing concerns the permission to use protected works in connection with an artificial intelligence system. The relevant activity may include collecting works for training, storing or processing them, supplying them as prompts, retrieving them through a search or retrieval system, and reproducing material in an output. Each activity requires separate consideration because permission to access a work does not necessarily permit copying, training, commercial exploitation or onward distribution.
Harper James advises UK businesses on commercial legal matters under the laws of England and Wales, including intellectual property and licensing arrangements. A business developing or deploying an AI system should identify the rights it needs, the works involved, the parties granting permission and the permitted purposes.
An AI licence should define the licensed material and permitted uses with precision. Important provisions include the duration and territory of the licence, whether use is exclusive, restrictions on training or fine-tuning, rules for subcontractors and cloud providers, confidentiality, security requirements, attribution, audit rights and termination. The agreement should also address whether the licensee may retain copies, create derivatives, combine the material with other datasets or use outputs commercially.
Rights ownership and warranties require particular attention. A licensor may warrant that it controls specified rights, but the contract should state the scope of that warranty and the consequences of a third-party claim. Indemnities, liability caps, procedures for handling infringement allegations and obligations to remove or replace disputed material determine how risk is allocated between the parties.
A licence for training data does not automatically give the business ownership of every output. Output ownership depends on the contract, the nature of the output and applicable copyright law. In the United Kingdom, copyright generally requires an identifiable qualifying work and human authorship. An output that reproduces protected expression may raise infringement issues even where the user does not own the underlying material.
A practical review starts with an inventory of datasets, prompts, models, plugins and outputs. The business can then record the source and licence for each material, restrict use where the licence is unclear, retain evidence of permissions, and establish a review process for outputs that resemble existing works or contain third-party content. This process should be reflected in supplier agreements, internal policies and customer terms where AI-generated material forms part of a commercial service.