As an innovative solution, synthetic data provides a secure and efficient alternative data source for AI training by algorithmically simulating real data’s statistical characteristics and distribution patterns. It reduces data collection costs, mitigates privacy leakage risks, and resolves copyright issues, gradually becoming a key driver of AI development. However, widespread use also brings complex governance challenges. During generation and use, sensitive personal information may leak, and the opacity of the process may exacerbate the algorithmic ‘black box’ problem, making decisions difficult to trace and explain. Meanwhile, large technology enterprises, leveraging first-mover advantages in data and technology, may further consolidate market dominance through synthetic data, triggering new forms of data monopoly. Low-quality or biased synthetic data may also cause models to disconnect from reality, leading to algorithmic discrimination or even system collapse. Facing these challenges, China urgently needs a governance system adapted to synthetic data. Future governance should take standardization, transparency, socialization, and high quality as core values, guiding institutional design that balances innovation and risk prevention by enhancing privacy protection, breaking algorithmic ‘black boxes’, preventing data monopolies, and avoiding data distortion. On this basis, synthetic data can be incorporated into data protection law frameworks with more relaxed standards, while relying on compliance audits, open-source software, and quality control to build a forward-looking governance system with Chinese characteristics.
内容提要 合成数据作为真实数据的安全替代品，在降低AI训练成本、规避隐私与版权侵权方面具有显著的优势。尽管如此，在合成数据训练过程中，依然存在泄露个人隐私、加剧算法“黑箱”、助长市场垄断以及导致算法决策与现实脱节等多重风险。合成数据在人工智能领域所引发的技术变革，对我国数据治理法律体系的完善提出了新的要求。未来合成数据的治理应当以规范化、透明化、社会化与高质量化为价值取向，推动形成兼顾技术创新与风险防范的治理框架。在制度建构上，可以将合成数据纳入现行法的规制范围，并为其提供适度宽松的隐私保护，同时依托合规审计、开源软件及质量管控等创新机制，逐步构建既面向未来又具有中国特色的合成数据治理体系。