INTRODUCTION Contributed by: Dale Cendali, Joshua Simmons and Dana DeVlieger, Kirkland & Ellis LLP
If the copyright holder opts out “in an appropri - ate manner”, then researchers need to obtain permission to use the work for TDM. At this point, it is unclear what constitutes opting out “in an appropriate manner”, but organisations like Pictoright (Netherlands) and Sacem (France) have issued public statements that use of their members’ works for TDM purposes requires prior authorisation. The use of such statements is expected to increase as the provisions of the Act take effect. In addition to implementing measures to avoid infringement, there are other considerations worth keeping in mind for AI developers, includ - ing: • getting legal access to the copyrighted con - tent used as training data; • checking for the existence of opt-out state - ments; • creating agreements with dataset owners and AI customers to divide liability for infringe - ment; and • limiting the use of training data content for other infringing purposes. Japan In Japan, as in much of the world, there have not been many laws or regulations passed or any judicial decisions regarding generative AI, although Japan did incorporate a TDM excep - tion in Article 30(4) of its Copyright Law in 2018. As AI capabilities have increased, this excep - tion has become increasingly relevant and Japan has responded by commissioning non-binding reports and guidelines addressing the challeng - es that generative AI presents for copyright law. The first such report is the “Report on AI and Copyright Issues”, published on 15 March 2024 by the Legal System Subcommittee of the Cul -
tural Council of Japan. It addresses three main topics: first, the risk of copyright infringement during the development of generative AI models; second, the risk of copyright infringement when using generative AI models; and third, the gov - erning law that applies to copyright infringement regarding generative AI. Infringement risk in development : While the Copyright Act of Japan does not have a gen - eralised fair use defence, it does have the Non- Enjoyment Purpose Requirement, which allows for use of a copyrighted work in any way where the purpose is not for the user to enjoy the work. The use cannot, however, unreasonably preju - dice the copyright holder’s interests. The test for whether a copyright holder’s interests are unrea - sonably prejudiced is whether the use conflicts with or hinders the market for the copyrighted work. The report addressed scenarios in which use of copyrighted works in the development of generative AI models hinders the market of the copyrighted work. The remedies for copyright infringement in the development of generative AI are injunctive relief and monetary damages. Infringement risk in use : Evaluation for copyright infringement in Japan is the same for AI out - puts as for traditional works – the infringing work must be similar to the copyrighted work and the infringing work must have relied on the copy - righted work. The report addresses two ways that an AI-generated work meets the reliance requirement and therefore infringes the copy - righted work – when the AI user recognises the copyrighted work, and when the AI user was unaware of the copyrighted work but the copy - righted work was included in the training data. On the other hand, if an AI user is not aware of the copyrighted work and the copyrighted work was not included in the training data, the reli - ance requirement is not likely to be met. When
7
CHAMBERS.COM
Powered by FlippingBook