Generative AI (GAI) is challenging copyright legislation in all jurisdictions. Key issues relate to the use of copyright material as training data, the originality and ownership of generated outputs, and the economic and social impacts on copyright-based industries.
GAI models are often trained on copyrighted material scraped from the internet without the consent or compensation of creators. This has led to several high-profile lawsuits, mainly in the U.S.
Proposed solutions include:
Some publishers have made licensing deals with AI developers, and authors may need to consent to AI training when signing publishing contracts. Ithaka S+R maintain a Generative AI Licensing Agreement Tracker.
Training on prompts raises major concerns in relation to users' privacy. Similarly, copyright may be infringed or commercial subscription agreements may be breached when users upload documents.
For example: uploading a journal article accessed through a university library may breach the university’s agreement with the journal publisher. For guidance on this issue see the Generative AI at university library guide.
Copyright law only protects human-created works. Purely AI-generated content is generally not eligible for copyright, however, the human-authored parts of a mixed work might be protected. Where an AI tool is used with significant human intellectual effort, the copyright status of the output becomes unclear and can vary by country, making international sharing more complicated.
This issue creates economic uncertainty for creators using GAI and for organisations using GAI for brand assets, advertising, or trademarks. In scholarly communication, it affects publisher editorial policies, who can grant reuse permissions, and what can be covered under open access licences.
There is a risk of infringement if an AI tool generates outputs that are substantially similar to an existing copyright work. Legal liability for the infringement may lie with the user, the AI developer or both, depending on the circumstances (Kluwer Copyright Blog).
Hallucinations from mainstream GAI tools occur less frequently than in 2023, however, GAI can still misinterpret content and summarise inaccurately. Misquoting or false attribution may infringe an author’s moral rights under copyright law.
Copyright influences the economic and social impacts of GAI because it determines who can legally control, licence, and profit from both training data and GAI outputs.
In Australia, there is no copyright exception for text and data mining, which may limit domestic AI development and increase reliance on offshore GAI tools. Both collective licensing arrangements and publisher licensing deals could increase costs for access to information for educators, students and researchers. Nevertheless, creators are entitled to fair compensation and protection of their work. Finding a fair and workable balance between innovation, access, and creator rights continues to evolve.