AI Tracker
Expanded to include a new benchmark exploring the policy response to the intersection of AI and copyright
We’ve added a new benchmark to our AI Tracker on the intersection of AI and copyright. The benchmark details legislative approaches to how copyright law is impacted by AI, focusing on the governance of the use of copyrighted works in AI training. Specifically, it highlights provisions in relation to rules on the use of copyrighted materials, how and if rightsholders’ works are protected, whether there are legal guarantees for rightsholder remuneration and how transparent AI firms must be with their training materials.
So far, the approaches we have in our benchmark mostly fall into the category of ‘opt-in’ or ‘opt-out’ regimes – the former meaning individual rightsholders can opt-in to their copyrighted works being used in AI training and the latter meaning they would have to actively opt-out to prohibit such use. The benchmark shows that four jurisdictions (EU, Japan, South Korea, UK) so far have adopted or proposed opt-out regimes, while only two (Brazil, Switzerland) are pursuing opt-in. Currently, only in Brazil has it been proposed that rightsholders will be remunerated for their work being used in the training of AI. Three jurisdictions (Brazil, EU, UK) require (or have proposed) that AI firms be more transparent about the data being used in their AI training. In both Brazil and the EU, firms would be required to publicise key information about the training data being used, while the UK’s proposed approach has more broadly suggested that a text and data mining exception should be underpinned by greater transparency requirements without yet specifying what those should be.
The benchmark also shows that a proposed amendment to the Copyright Act in Denmark would mean the country would become the only one that does not provide any legally binding options for rightsholders to protect their works from being used in the training of AI. The amendment instead focuses on protecting citizens from uses of their image or likeness in AI-generated deepfakes by prohibiting digital imitations of individual characteristics, artistic works, literature and performances until 50 years after the death of the relevant artist or individual.
