Control and Compensation. A Comparative Analysis of Copyright Exceptions for Training Generative AI

Abstract Lawmakers and administrative agencies around the globe are debating whether the use of copyrighted content for AI training does or should require the rights holder’s consent. This article examines legislation and policy debates in the U.S., Canada, the UK, the EU, Israel, China, Singapore, and Japan. Issues of control, compensation, transparency and legal certainty dominate the discussion. Countries are trying to recalibrate the balance of interests, either in favour of AI companies, or by supporting rights holders – for example, with copyright-related transparency obligations. EU copyright law offers a relatively favourable environment for AI companies. Ultimately, however, copyright law is not the decisive factor when AI companies choose the location of their training facilities.