关键词: 开源AI治理, 标准话语权, 数据跨境流动, AI物料清单, 数据科学

Abstract:

The imbalance of standard discursive power in open-source AI governance hinders the transformation of China’s de facto advantages in open-source models into institutional discourse power. Taking standard discourse power as its analytical object, this paper examines the manifestations and causes of this imbalance: the board of the Open Source Initiative (OSI) has no Asian representatives, its open source AI definition (OSAID) has been widely criticized for not requiring the disclosure of training data, and the open-source exemptions of the EU AI Act are not applicable to high-risk scenarios. Meanwhile, Chinese open-source AI models have become de facto standards, with Qwen generating over 113000 derivative models and DeepSeek ranking first in attention on HuggingFace, yet there remains an evident gap between institutional discourse power and actual contributions. This paper analyzes the power dynamics behind standard competition, examines the institutional dilemmas of cross-border data governance, proposes pathways for building an autonomous standard ecosystem through AI bill of materials (AIBOM) standardization, trusted data spaces, and regional compliance mutual recognition, and discusses the knowledge foundation and talent support for standard building from the perspective of data discipline construction.

Key words: open source AI governance, standard discourse power, cross-border data flow, AI bill of materials, data science