English home / AI, training & text and data mining
AI, training & text and data mining
Review input data, training, model use and published output as separate legal stages.
German copyright law and relevant European rules. The appropriate solution depends on the facts, agreements and intended use.
Legal framework
AI projects can involve copying and processing protected material as well as generating and publishing outputs. Publicly accessible content is not automatically available for every processing purpose. Establish the origin of training data, lawful access and the specific technical activity before relying on an exception.
Practical guidance
Section 44b of the German Copyright Act addresses reproduction for text and data mining of lawfully accessible works, subject to its requirements, including effective rights reservations. For works available online, the reservation must be machine-readable. Scientific research is subject to distinct provisions. Whether a particular AI operation falls within an exception requires assessment rather than a blanket assumption.
Purely machine-generated output is not automatically protected because a business paid for the service. Human creative contributions should be assessed and documented. Even unprotected output can infringe third-party rights. Provider agreements should address allowed inputs, reuse of data, confidentiality, claims handling and indemnities, with a human approval process for important publications.
Practical examples
These fictional examples illustrate common issues; they are not accounts of particular client matters.
Example 1
Internal assistant using journal articles
A company loads subscription texts into a knowledge system. Reading access does not automatically authorise every storage, processing or output activity. Review the licence, technical workflow and staff access.
Example 2
AI image promised as exclusive branding
An agency offers exclusivity in a generated image. Assess the human contribution and the provider’s commitments before promising a legal position that may not exist.
Example 3
Publisher reserves mining rights
A publisher wants to restrict mining of online content. The rights held, machine-readable implementation and timing must be documented together. Technical marking and legal effect require assessment in the actual scenario.
Frequently asked questions
Does the customer automatically own AI output?
Not in a single, universal sense. Copyright protection, contractual permission and third-party rights are different questions.
Is an ordinary website notice enough for a mining reservation?
For online works, section 44b requires machine readability. Do not assume any general notice satisfies the specific requirements.
Preparing for advice
01 · Your objective
Explain the intended use or dispute, the parties, countries and any deadlines.
02 · Your documents
Gather agreements, licence terms, drafts, evidence of use and relevant correspondence.
03 · Next steps
We agree the scope and fees, then assess the permissions, contractual changes or proceedings required.
Related topics
Databases & scientific works · Software licensing & open source · Copyright licences & exploitation rights
Advice from German specialist lawyers
Michael Horak and Julia Ziegeler are German-certified specialist lawyers in copyright and media law. Hanover is our headquarters. We advise at our offices and by video conference. Meet the team · Contact us in English.
Legal reference: German Copyright Act (official German text).