Legislative Acts

Paid Licensing for AI Training Data

Taiwan’s books may be turning into training data for foreign AI. Meanwhile, the government’s own sovereign AI effort is asking authors to license their work for free. On November 4, 2026, Legislator Ju-Chun Ko questioned Minister of Culture Li Yuan in the Legislative Yuan’s Education and Culture Committee, demanding that the government trace where Taiwanese books are going, set up compensation for creators, and build a legal, fair, and paid channel for licensing AI training data.

Where did 563 tonnes of Taiwanese books go?

The international publishing world has been watching Project Panama, an effort by U.S. AI company Anthropic to buy physical books in bulk, take them apart, scan them into digital text, and use the text to train AI models.

According to South Korean media reports, Taiwanese book distributors have shipped about 563 tonnes of books to the United States in 16 batches since November 2024. That could be as many as 1.88 million books, equal to 36% of the physical collection of Taiwan’s National Central Library. The buyers were used-book dealers linked to Anthropic’s book acquisition program. Whether these books were also used to train AI deserves a closer look.

Ko’s question is simple: if foreign AI companies train on Taiwan’s books, are Taiwanese creators being paid fairly? And what about Taiwan’s own AI?

Foreign AI learns Taiwan while Taiwan’s AI can’t read Taiwan’s books

Traditional Chinese text and Taiwan’s history and culture are essential data for AI that understands Taiwan and answers Taiwanese users well. The problem: foreign AI companies can now buy Taiwanese books in bulk, while domestic AI developers who want to pay for training data legally keep hitting walls because copyright rules are unclear and licensing channels are scarce.

The likely result: foreign AI gets better and better at understanding Taiwan, while Taiwan’s own AI can’t read Taiwan’s books.

Sovereign AI corpora can’t run on “free licensing”

The Ministry of Digital Affairs (MODA) recently launched a call for Taiwan sovereign AI corpus contributions, and it encouraged publishers and e-book platforms to license their content for free. Creators spend years on their work, yet the government, in the name of AI development, simply holds out its hand for free material. Ko asked: should national technology policy be run like a “cultural beggar”?

Minister Li Yuan answered clearly: “The Ministry of Culture’s position cannot be free!” He added that the government should set up a compensation mechanism and fund protections for creators’ rights.

Not the first time: the corpus licensing problem is long overdue

Sovereign AI needs Taiwanese text, and that text needs legal licensing. Ko has been pressing this point since the 2025 review of the AI Basic Act:

1. April 29, 2025, general inquiry (education session): Ko raised a series of questions on open data, sovereign AI, and AI training databases.
2. May 1, 2025, Legislative Yuan inquiry: Ko proposed that the Executive Yuan coordinate a national AI corpus database and set up an open-source office. MODA replied that it had 50,000 open datasets and was drafting licensing terms for a sovereign AI corpus with the Intellectual Property Office.
3. May 14, 2025, joint review of the AI Basic Act: Ko asked how the government would obtain data from the private sector. MODA Deputy Minister Yeh Ning admitted that corpus collection had no payment mechanism and relied on voluntary contributions and rewards; MODA was also seeking NT$40 million from the NSTC to build the corpus.

A year and a half later, licensing terms have been discussed and budgets requested, yet in 2026 the government is still asking authors to license their work for free. Foreign AI companies have already paid to haul off Taiwan’s books by the shipload, while Taiwan’s government is still stuck on “voluntary” and “rewards.”

Ko’s demands

1. Actively trace the books: On foreign AI companies buying Taiwanese books in bulk, the Ministry of Culture is still mostly “searching and observing.” Political Deputy Minister Lee Ching-hwi said the ministry will keep following related U.S. lawsuits, and Li said that when needed the ministry will help authors defend their rights overseas. The ministry can’t just wait. It should find out where the books went and how they are used, and when needed help publishers and authors defend their rights.
2. Compensate creators: The corpora the government needs for AI should come with a budget and fair payment to authors, not requests for free licenses.
3. Build an AI training data licensing platform: The Taiwan Creative Content Agency (TAICCA) or another relevant agency should build a platform that connects AI companies willing to pay with publishers and creators willing to license. Minister Li promised feasibility findings within one month.
4. Clarify copyright rules: Domestic AI developers who want to pay for training data legally need clear copyright rules. Minister Li agreed to provide amendment input within one week.

Next steps

Taiwan’s books are not just products. They carry our language, culture, and knowledge. Ko will keep pressing the Ministry of Culture, the Ministry of Economic Affairs, and MODA to work together on a legal, fair, and paid system for using AI training data, so creators are protected, publishers gain a new market, and Taiwan’s AI truly understands Taiwan and passes this cultural heritage on to the next generation.

Related pages: AI Basic Act, AI Equity and AI Voucher Initiative.

Timeline