As generative AI emerges as a pivotal force in today's digital age, it prompts an earnest reevaluation of established copyright tenets. The convergence of AI and the law is not merely a theoretical challenge; it's a pragmatic shift in how we understand creativity and ownership in the digital realm. While some legal aspects fit neatly into existing frameworks, others raise unprecedented issues, forcing us to find new ways and possibly rethink fundamental principles.
The following core issues arise along the user process of generative AI:
Training: Can copyrighted works like texts or images be used to train an AI model?
Under Sec. 44b German Copyright Act, implementing Art. 4 of the EU Copyright Directive, copyrighted works that are lawfully accessible can be replicated for training purposes in Germany. This permission stands as long as the purpose of this text or data mining remains training. However, this rule is negated if the copyright holder expresses a machine-readable reservation (e.g., in the robots.txt).
Input: What are the copyright implications when using copyrighted works as input to an AI tool?
If a user copies copyrighted works of third parties into their AI Tool's prompt, it may, depending on the case, be an unauthorised duplication requiring permission. This may result in claims for cease and desist from third parties, provided they become aware of it.
If the prompt demonstrates a degree of originality (and length), it might itself establish copyright protection. This can represent a valuable asset for specific businesses (e.g., licensing packages of valuable prompts).
Output: Can the output of an AI tool infringe on others' copyrights or be copyrighted itself?
AI output can potentially breach third-party copyrights if it contains copyrighted content (e.g., text passages, pictures, melodies, et alt.). The (commercial) use of such compromised AI part outputs may expose users to legal risks, particularly cease and desist claims.
Output: Can the output of an AI be copyrighted itself?
The output from AI is not inherently copyrighted. German and European copyright laws uphold the principle that a human must be the creator of a specific work (see also CJEU’s judgement dated of 12 September 2019, case C-683/17 – Cofemel). Often, AI determines the final look of the work, and the users’ prompts can still allow AI to generate thousands of possible designs. For this reason, the AI ultimately determines the design of the output based on the prompt. However, an exception might arise in iterative processes where humans extensively modify and reshape an AI's output, using it as a first draft.