AI development can involve protected text, images, software, recordings and databases. Lawful access, the relevant text-and-data-mining exception, effective rights reservations and contractual restrictions all need analysis. Section 44b of the German Copyright Act provides a framework for certain mining uses; it is not an unrestricted licence to collect and reuse everything online. Research-specific rules require separate consideration.
How we approach the matter
We examine dataset provenance, rights information, access conditions and the proposed technical uses. Licence negotiations should distinguish training, fine-tuning, retrieval, evaluation and redistribution. Output review remains necessary where results reproduce protected material or affect personality rights. Purely machine-generated output and human creative contributions must be distinguished when assessing protection and exploitation rights.
From classification to operational controls
Describe the actual use case before selecting a compliance checklist: the inputs, outputs, decision-making process and affected people all matter. A supplier’s product category does not settle the customer’s legal role. Allocate responsibility for documentation, human review, transparency and incident escalation. Data protection, confidentiality, copyright and sector-specific rules continue to apply alongside the AI Act.
Procurement, change and evidence
Contracts should secure the information and assistance needed to operate the system lawfully. Assess how model updates, new data sources or a change of purpose will be reviewed. Keep a record of the approved use, limitations and relevant tests; this is more useful than a general statement that the product is compliant. If a concern arises, preserve the configuration and relevant input-output examples, subject to appropriate confidentiality and personal-data safeguards.
Four practical examples
The following scenarios illustrate typical legal questions. They are not descriptions of client mandates or promised outcomes.
Publisher dataset
A developer wants to train on a news archive. We review permissions, rights reservations and alternatives to unlicensed collection.
Image output
An agency receives an output resembling a known photograph. We review the similarity and proposed use before recommending clearance or replacement.
Retrieval service
A business indexes licensed documents for a retrieval system. The review checks whether storage, snippets and customer access fall within the licence.
Training restriction
A rights owner wants to reserve mining uses. We assess appropriate machine-readable measures and how they interact with existing contracts.
Preparing your enquiry
A short description of the project or dispute, the parties involved, relevant countries and any deadline helps us identify the next step. Preserve the relevant contract, publication or technical records. Please avoid sending highly sensitive material until a suitable communication channel has been agreed.
Discuss your matter with our team · Media law overview · Disputes and enforcement