The European legal framework regulating the relationship between Artificial Intelligence (AI) and intellectual property and related rights is based on the combined application of two central pieces of legislation: the Directive on Intellectual Property Rights in the Digital Single Market (DSM Directive 2019/790) and the recent Artificial Intelligence Act (AI Act - EU Regulation 2024/1689).
This framework is divided into two main axes: the stage of data input for training models (Input) and the stage of protection of works produced by AI (Output).
1. AI Model Training (Input) & Text and Data Mining (TDM)
The use of protected works for training algorithms is regulated by Directive (EU) 2019/790 through the exceptions for Text and Data Mining (TDM):
- Scientific Research Exception (Article 3): Data mining by research organisations and cultural heritage institutions for the purposes of scientific research is freely permitted.
- General Exception / Commercial Use (Article 4): Anyone (including AI companies) is permitted to carry out TDM, unless the rightholders have explicitly exercised the right to opt-out.
- Opt-out mechanism: The declaration of the authors’ opt-out must be made by appropriate means that are readable by machines (e.g. via the robots.txt file on a website).
2. AI Providers’ Obligations under the AI Act (Regulation 2024/1689)
The AI Act introduces strict horizontal obligations, especially for providers of General-Purpose AI (GPAI) models, such as Generative AI models:
- Compliance Policy (Article 53(1)(c): Providers are required to implement a policy of respecting European copyright law. They must identify and comply with the opt-out statements of creators, regardless of the country in which the model is trained (extraterritorial validity).
- Transparency Obligation (Article 53 par. 1 f. d): Companies must compile and publish a detailed summary of the content (data) used to train their model, based on a standard issued by the European AI Office.
- Content Marking (Article 50): There is an obligation to clearly mark (watermark) that the produced material (text, audio, video, image/deepfakes) is a product of artificial intelligence.
3. Protection of Works Produced by AI (Output)
Regarding whether the output of AI itself can be protected by copyright: [1]
- Human Creator: According to the settled case law of the Court of Justice of the European Union (CJEU - e.g. Infopaq decision), protection presupposes “personal intellectual creation”.
- Exclusion of Automated Works: Works produced entirely automatically by an AI (without human intervention) are not protected by copyright law.
- Hybrid Creation: If the AI is used merely as a supporting tool and the human retains creative control (e.g. extensive parameterization, selection, editing), the final work can be protected.
Current Developments
- From August 2025: GPAI model providers must publish the summaries of the training data (template).
- In March 2026: The European Parliament adopted a resolution to create a balanced framework that protects creators and enables innovation, signaling a possible shift towards stricter measures (Axel Voss Report).
- AI Works: According to the case law of the CJEU, works produced exclusively by AI are not protected, unless there is significant human input.