Unauthorized Central News Agency Content in AI Training Dataset
A Taiwanese PhD student shared an open-source Traditional Chinese dataset containing approximately 140,000 unauthorized news articles from the Central News Agency in a Facebook AI group. The Central News Agency filed a criminal complaint, and authorities are investigating the incident.
What happened
In March 2025, a person shared an open-source Traditional Chinese dataset called fineweb-zhtw in the Facebook group 'Generative AI 技術交流中心', made available for public viewing, modification and distribution for large language model training.
The dataset contained approximately 140,000 news articles from the Central News Agency (中央社) published between 2011 and 2021, without authorization from the news organization.
Reported impact
The Central News Agency described the unauthorized use of its news content as compromising its copyright and business interests, stating that news content represents significant assets for media cooperation and licensing.
Index analysis
The incident occurs in the context of a recent high-profile case where Lawsnote (七法公司) was convicted of copyright infringement for scraping a competitor's legal database, raising questions about data use boundaries in AI development.
Governance controls
The Digital Development Ministry stated that web scraping itself is not prohibited, but highlighted concerns about scraping privately held databases versus authorized government datasets, and encouraged AI companies to use government-provided training corpora instead of private enterprise data.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.