Open-Source AI: Definition, Risk and Governance
- Time
- 2026-08-08 10:20 ~ 10:50
- Speaker
- Cui Jia Wei
- Room
- TR411
- Co-write
Abstract
Since DeepSeek burst into the spotlight during the 2025 Lunar New Year, discussions around open-source artificial intelligence have moved from technical and open-source communities into the public eye. However, the relationship between open-source communities and the development of AI runs far deeper than the recent attention triggered by DeepSeek. Many of the training frameworks that underpin today’s AI technologies have been released as open source, and the origins of both the open-source and free software movements are closely tied to the MIT AI Lab. More recently, open-source AI has also become a focal point for national attention due to the rise of “sovereign AI” debates. Against this backdrop, this session will focus on the definition of open-source AI, its associated risks, and a comparative analysis of national policies. First, what exactly counts as “open-source AI”? Is releasing model code alone sufficient to qualify as open source? Should model weights, or even the training datasets, also be made publicly available? And are the so-called “open-source” AI systems promoted by major tech companies truly open? Second, given the ease with which open-source AI can be disseminated, does it introduce new risks compared to closed-source systems? Or, alternatively, can the transparency enabled by open source serve as a tool for risk management and mitigation? Third, how are different countries approaching open-source AI? The European Union, through the EU Artificial Intelligence Act, explicitly provides that AI components released in a free and open-source manner may be exempt from certain regulatory requirements. However, are there exceptions or limitations to this exemption? What is the stance of the United States government, particularly as technological competition between the U.S. and China intensifies, and China takes a leading role in releasing multiple open-source AI models? How is Singapore positioning its policies toward open-source AI? Finally, what lessons can Taiwan draw from these global discussions?
Speaker
Cui Jia Wei
崔家瑋目前就讀於台灣大學法律研究所，研究領域為開源人工智慧治理。他於 2022 年通過律師與司法官考試，在台灣最大的虛擬資產交易所擔任法律專員。在學術與工作之外，他亦參與台灣的非政府組織與公民科技社群。他所參與的公民科技專案 vTaiwan 探討數位工具應用於審議民主的可能性。vTaiwan 在 2023 年與 OpenAI 合作，探討利用審議民主改善人工智慧治理的提案；自 2024 年開始，vTaiwan 與台灣網路資訊中心合作，以數位工具協助網路與人工智慧治理的多方利害關係人討論。他於 2025 年亦獲選成為 TWNIC Academy Fellow。 Cui Jia-Wei is currently a graduate student at the Graduate Institute of Law at National Taiwan University, where his research focuses on the governance of open-source artificial intelligence. He passed both the bar examination and the judicial officer examination in 2022, and currently works as a legal specialist at Taiwan’s largest virtual asset exchange.
In addition to his academic and professional work, he is actively involved in non-governmental organizations and civic tech communities in Taiwan. He participates in the civic tech project vTaiwan, which explores the use of digital tools in deliberative democracy. In 2023, vTaiwan collaborated with OpenAI to develop proposals on improving AI governance through deliberative democracy. Since 2024, vTaiwan has partnered with the Taiwan Network Information Center to facilitate multi-stakeholder discussions on internet and AI governance using digital tools. In 2025, he was selected as a TWNIC Academy Fellow.