> ## Content Index
> Fetch the complete content index at: https://nordisk-ai.ghost.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# Harvard Launches Massive AI Training Dataset With Support From OpenAI and Microsoft
- URL: https://nordisk-ai.ghost.io/harvard-launches-massive-ai-training-dataset-with-support-from-openai-and-microsoft/
- Published: 2025-02-07T14:00:35.000Z
- Updated: 2026-08-27T15:08:24.000Z
- Description: Harvard is making AI development more accessible by releasing a dataset of a million public-domain books. Will this change the game for AI training and break the reliance on copyrighted material?
- Author: Anders Hagen
- Tags: AI News, AI, AI training, aisweden, article, dataset, Harvard, Microsoft, nordiskai, OpenAI, public domain, copyright law, #Migrated-1787841634512, #wp, #wp-post, #Import 2026-08-27 14:41, #Import 2026-09-07 13:41

Harvard University has unveiled a dataset of nearly a million public-domain books, now available for training AI models. The creation of the dataset is part of Harvard's new **Institutional Data Initiative (IDI)** and has been funded by Microsoft and OpenAI. The books, originally scanned through the Google Books project, span multiple genres, time periods, and languages. Classics by Shakespeare and Dante sit alongside obscure Czech math textbooks and Welsh pocket dictionaries.

#### A Step Toward Greater AI Accessibility

The dataset, which is five times larger than the widely discussed [**Books3 dataset**](https://paperswithcode.com/dataset/books3?ref=nordisk-ai.ghost.io), aims to democratize access to high-quality material for AI training. According to Greg Leppert, executive director of IDI, it should level the playing field by giving smaller players and researchers access to the same kind of resources big tech companies have. "It has undergone rigorous review," says Leppert, noting that it can be combined with licensed material to build competitive AI models.

#### Support From Microsoft and OpenAI

Microsoft's deputy general counsel for intellectual property, Burton Davis, emphasizes that the project aligns with the company's ambition to create accessible data sources for AI development. OpenAI has expressed similar enthusiasm, describing the support as part of its long-term strategy to back ethical, open AI initiatives.

#### A Growing Trend for Open Datasets

The Harvard initiative is just one of many projects producing publicly available datasets. Other examples include the French project **Common Corpus**, which contains 3-4 million books and periodical collections, as well as initiatives like **Source.Plus**, an image database created by AI startup Spawning. These projects aim to offer high-quality resources that don't risk legal problems related to copyright.

#### The Debate Over AI and Copyright

As these datasets are launched, legal battles over the use of copyrighted material for AI training continue. Some argue that large open databases like Harvard's undermine the arguments from AI companies claiming that copyrighted material is indispensable for creating high-performing models.