Concerns over artificial intelligence have surfaced across industries, and a recent incident involving a major technology firm has raised questions about intellectual property practices. A tutorial published on Microsoft's developer platform utilized what seems to be an illegally obtained collection of J.K. Rowling's Harry Potter novels to illustrate training an AI tool on the Azure cloud service.
In the entry, Pooja Kamath, a senior product manager at Microsoft, introduced the series by noting that it comprises seven volumes chronicling the experiences of young wizard Harry Potter and his companions as they confront the malevolent Voldemort and his followers. The article referenced a dataset on Kaggle featuring seven text files that collectively hold the full content of the published books.
The tutorial focused on integrating generative AI capabilities into software applications using Azure. Kamath highlighted potential applications such as building question-answering tools or producing fan-generated tales in the Harry Potter universe, describing it as an engaging option for enthusiasts to invent fresh narratives and embark on imaginative quests. The piece concluded with an AI-created illustration depicting two youngsters aboard a train, resembling Harry Potter and Ron Weasley, flanked by the Microsoft emblem.
Such actions represent a significant violation of intellectual property rights. The Harry Potter books remain protected by copyrights owned by multiple parties worldwide, including the author. Current listings on Amazon indicate that the e-book bundle retails for 70 dollars. Distributing or acquiring these materials without compensation to rights holders constitutes infringement in most jurisdictions, regardless of the purpose, including use in machine learning models.
The Microsoft guide appeared in late 2024 and was subsequently taken down from the official site, though copies persist on the Internet Archive. The associated Kaggle dataset had been incorrectly labeled as public domain and garnered around 10,000 downloads, as detailed in an Ars Technica analysis. The materials evaded widespread notice for approximately 18 months until a discussion on Hacker News recently spotlighted them.
The incident underscores a lax approach to digital content rights in an official company resource, potentially stemming from a misunderstanding of public domain classifications. However, many leading AI systems have undergone training on vast libraries of electronic books, a substantial portion of which originated from unauthorized sources. Legal challenges from authors targeting companies like Meta, OpenAI, Nvidia, Google, Anthropic, and Microsoft seek to halt such training practices or demand compensation for unlicensed usage, with court outcomes varying—some rulings deem the AI outputs as transformative under fair use doctrines, while others uphold the need to address initial unauthorized access.