Huawei and DeepSeek Linked to Pirated Training Data
Researchers find that major AI models were trained on pirated books, including works by multiple Korean authors.
Updated on Oct. 26, 2026 in Artificial Intelligence
Live Poll
Should technology companies be required to obtain explicit permission to use copyrighted books for AI training?
Huawei and DeepSeek utilized datasets containing pirated e-books to train their artificial intelligence models. Research indicates these unauthorized collections include English translations of works by Korean authors.
Why it matters
The use of pirated digital content allows AI developers to improve model performance in narrative coherence and long-context processing. This reliance on unauthorized materials raises significant copyright concerns within the global tech industry.
DeepSeek utilized a massive repository consisting of 860,000 English-language e-books and 180,000 Chinese-language e-books from Anna's Archive to train its DeepSeek-VL model. Meanwhile, Huawei incorporated the 800-gigabyte Pile dataset, which contains the Books3 collection of pirated texts.
The players
Huawei
This Chinese multinational technology corporation specializes in telecommunications equipment and consumer electronics, including the development of AI models like Pangu Sigma.
DeepSeek
This China-based AI development firm focuses on creating large language models and vision-language architectures.
Han Kang
She is a celebrated South Korean author whose literary works, including The Vegetarian and Human Acts, have been identified in pirated AI training datasets.
Anthropic
This American AI safety and research company notably settled a copyright lawsuit by agreeing to pay $3,000 per book for using unauthorized works.
The details
Researchers found that the Books3 collection, embedded within The Pile, houses at least 19 books written by 10 Korean authors. Furthermore, tests revealed DeepSeek 3.1 successfully reproduced 19 consecutive words from the novel 'The Vegetarian' by Han Kang.
Timeline
1996: Kim Young-ha's I Have the Right to Destroy Myself is published.
2007: Han Kang's The Vegetarian is published.
2008: Shin Kyung-sook's Please Look After Mom is published.
2014: Han Kang's Human Acts is published.
2022: Zhipu AI and Tsinghua University report is published.
The Tech Race
The current allegations against Huawei and DeepSeek follow the pattern set by the Anthropic copyright lawsuit settlement regarding the use of unauthorized books. This situation highlights the ongoing legal tension as developers prioritize massive data intake to gain a competitive edge in AI capability.
Users interacting with advanced AI models should be aware that the content generated may be derived from unauthorized or copyrighted sources. This reliance on pirated data may lead to increased scrutiny, potential service disruptions, or future changes to how models are trained and licensed.
The takeaway
The unauthorized use of literature for model training highlights the struggle to balance rapid AI development with intellectual property rights. Companies continue to face pressure to secure cleaner training data as legal precedents regarding copyright infringement evolve.
Further reading
Learn more about the evolving standards for data acquisition in Artificial Intelligence.
Source note: This article includes information reported by Koreajoongangdaily.
Live Poll
Should technology companies be required to obtain explicit permission to use copyrighted books for AI training?
