Between 2018 and 2019, an employee of legal database A allegedly used six web crawlers to repeatedly and automatically replicate and download tens of thousands of electronic records, which included legislative histories, legal provisions, and attachments, from the legal database B, and incorporated such records into the legal database maintained by legal database A. In June, 2025, the New Taipei District Court rendered a judgment that the representative and employee of legal database A committed the offense of infringing property rights under Article 91, Paragraph 2 of the Copyright Act, as well as the offense of unlawfully obtaining computer electronic records under Article 359 of the Criminal Code. The judgment has sparked significant discussion within both the legal community and the technology industry.

In this judgment, the court held that legal database B owned the copyright in the editorial work, rejected legal database A’s fair use defense, referred to the U.S. federal court decision in Reuters v. Ross Intelligence (Ross Intelligence was found to have used content from the Westlaw database owned by Reuters to train its AI model without Reuters’ authorization, which was not deemed fair use) made in February 2025, and stated in the reasoning of the judgment that “referring to the criminal facts in this case, not every act of data acquisition can be excused under the pretext of ‘AI training or learning’ as a means to evade legal restrictions and sanctions.” The court further stated that this case did not even involve generative AI training; rather, it concerned the direct appropriation of copyrighted works for use in its own commercial legal database, an act the court considered more direct and egregious than the use of data for generative AI training.

It is noteworthy that although the facts of this case are clearly irrelevant to the training and learning of generative AI models, the court, while expressly rejecting legal database A’s fair use defense, still cited recent U.S. copyright infringement cases involving similar uses of database content. By distinguishing the “lighter” offense (using others’ database content for AI training) from the “heavier” offense (directly appropriating such content for one’s own database), the court concluded that the use in this case did not constitute fair use, marking Taiwan’s first ruling to explicitly refer to the applicability of fair use to AI training. A line from the court’s ruling “not every act of data acquisition can be excused under the pretext of ‘AI training or learning’” not only underscores the reproachability of legal database A’s conduct but may also influence future cases involving the use of others’ copyrighted works for AI training. However, how such use shall be assessed remains an open question, as this case does not involve AI learning or training, and the judgment therefore does not provide a detailed response on that issue. In Reuters v. Ross Intelligence, the U.S. federal court found in favor of the copyright holder with respect to two key fair use factors: the “the use’s purpose and character” and the “how the use affected the copyrighted work’s value or potential market.” These two factors were also given the greatest weight in the court’s ultimate analysis, leading to the conclusion that Ross’s use did not constitute fair use. Whether this approach may influence future determinations of the relative weight of fair use factors under Taiwanese law in similar cases remains to be seen.

Generative AI training data often includes materials that are part of the public domain or can be obtained through legitimate licensing channels, but such sources are inevitably limited, and challenges remain where holders of exclusive data restrict its use or refuse to grant licenses. Should the law prioritize accelerating the development of AI technologies or preserving the rights and interests of copyright holders? This remains an unresolved and complex question. In the pre-publication report Copyright and Artificial Intelligence Part 3: Generative AI Training, released in May 2025, the U.S. Copyright Office explicitly stated that “the public interest requires striking an effective balance, allowing technological innovation to flourish while maintaining a thriving creative community.” With legal database A having announced its intention to appeal, the development of the legal database case is certain to attract attention from Taiwan’s technology industry. Beyond the facts of this case, the broader question remains: can and how can the development of AI technology coexist with the doctrine of fair use? Countries around the world are watching closely.

(The article is originally in Chinese which can be found here.)