The draft includes several new obligations for AI providers, such as appointing a contact person for rightsholders and implementing safeguards against copyright-related overfitting, where AI-generated outputs closely resemble copyrighted input materials. However, the measures also contain notable caveats. For instance, AI providers can adopt measures to address ‘frequent unfounded complaints’ from rightsholders.
While the draft introduces a requirement for copyright due diligence when acquiring third-party datasets, the language remains nonbinding. Providers are asked to make ‘reasonable and proportionate’ efforts to ensure datasets comply with copyright law but are not explicitly required to verify the datasets themselves.
Several concerns have been raised by Culture Action Europe regarding the draft. One issue is that the proposed KPIs focus on formalistic measures rather than outcomes, potentially limiting their effectiveness. Additionally, many obligations do not apply to small and medium-sized enterprises (SMEs), which could create loopholes for certain businesses. The draft does not cover significant issues such as clear guidance on handling datasets containing infringing content, deepfake labelling, and strategies for compensating creators for the use of their intellectual property. Plans for removing infringing content from datasets remain absent as well.