Singapore’s legal framework permits the use of lawfully obtained works for artificial intelligence (AI) training, providing clarity for AI developers and avoiding contentious practices seen elsewhere, according to legal experts. This approach contrasts with recent international controversies in which companies reportedly purchased, scanned, and destroyed physical books—including rare editions—to feed AI models under the claim of fair use.
Legal specialists highlight that Singapore’s explicit provision allowing AI training on legally accessed content offers businesses much-needed certainty. Tony Yeo, vice-president of the International Association for the Protection of Intellectual Property’s Singapore Group and director of intellectual property at law firm Drew and Napier, said this clarity positions Singapore as a leader in AI and intellectual property law.
The remarks came during the Global Forum on Intellectual Property held at Marina Bay Sands on August 26, coinciding with the launch of a public consultation by Singapore’s Ministry of Law and Intellectual Property Office. The consultation seeks input on the impact of AI on copyright and patent laws and considers whether copyrighted materials can be used to train AI models.
Uncertainty in other jurisdictions has driven some technology companies to resort to destructive methods, such as slicing pages from printed books before scanning and discarding them. One U.S. court has deemed such scanning by Anthropic, a company behind the Claude AI chatbot, as transformative and fair use, but the practice has nonetheless sparked public backlash. Investigations have revealed shipments of rare books purchased by tech firms destined for such destructive scanning at Amazon facilities.
Experts note that Singapore’s 2021 amendments to the Copyright Act clarify that AI training on copyrighted works is lawful, provided the material is obtained legitimately. This means that using a purchased book or photograph for AI training is permissible within Singapore’s jurisdiction.
Kent Walker, president of global affairs at Google, emphasized the importance of legal certainty for fostering innovation in AI technology, warning that requiring developers to secure separate commercial licenses for every piece of data used would hinder progress. He cited Singapore, Japan, and the European Union as examples of regions with clear regulatory frameworks regarding AI training on publicly available content.
Legal practitioners also interpret these regulations to allow the use of freely accessible online materials, including website text and social media images, for AI training. One expert noted that rights holders cannot override this allowance simply by asserting that their published works should be excluded from AI training.
However, the law does not permit AI to generate output that infringes on copyrights, such as recreating images that are substantially similar or identical to copyrighted works, according to Yeo. Rights holders wishing to restrict use of their content for AI training may limit public access by placing works behind paywalls or disabling downloads. Yet, this approach involves trade-offs, as noted by Bird & Bird partner Oh Pin-Ping, who warned that reducing public visibility may diminish an author’s or artist’s audience.
For smaller rights holders, including individual creators, collective licensing organizations could offer a viable path to managing AI training licenses and ensuring fair compensation—a model already used in the United States.
Advocates for rights holders’ control stress the importance of allowing individuals to opt out of AI training use if they wish. Google’s Walker described opt-out mechanisms as a traditional, effective method for managing rights online, which could foster an open market where content creators can choose their level of participation in AI development.