The Research Institute for Democracy, Society and Emerging Technology (DSET) on Oct. 31 was invited to a closed-door roundtable hosted by the Global Taiwan Institute (GTI). At the forum, DSET’s Democratic Governance Program presented two forthcoming reports and engaged in in-depth discussion with experts from the D.C. think tank community. Participants included representatives from the Center for Security and Emerging Technology (CSET), Jamestown Foundation, Bertelsmann Foundation, Human Rights Foundation, Atlantic Council, and Doublethink Lab.
Kai-Shen Huang, DSET Director of the Democratic Governance Program, presented findings on China’s AI export strategies. He described China’s “application-driven” model, where AI products are scaled domestically before export. Huang also highlighted obstacles for U.S. data review, citing the opacity of China’s Personal Information Protection Law, Data Security Law, and Cybersecurity Law.
Global discussions on AI governance often focus on the race to develop frontier models, overlooking the global expansion potential of China’s application-driven development model. Through large-scale experimentation and feedback in areas such as smart cities, healthcare, and transportation, China’s AI industry has built a self-reinforcing cycle of technological optimization and commercial export. At the same time, state policies actively promote the overseas expansion of AI services, exporting not only technologies but also the governance norms embedded within them.
This “tight-at-home, open-abroad” dual structure makes Chinese AI services both highly competitive and politically sensitive in global markets. For democracies, the key challenge is to maintain openness and innovation while establishing effective import review and oversight mechanisms to guard against systemic risks and excessive dependence.
You-Hao Lai, Deputy Director of DSET’s Democratic Governance Program, presented key findings from his forthcoming policy report, “The Authoritarian Gaze: China’s Global Data Reach and Systemic Risks to Democracy.” He emphasized that democratic nations must not only focus on domestic AI development, but also remain vigilant against the infiltration of “authoritarian AI.”
Lai analyzed the authoritarian features of China’s data governance model, warning that it facilitates a one-way flow of data—“all in, rarely out”—that could reshape the global data order. He further argued that the data practices of Chinese AI services are reinforcing the global expansion of China’s authoritarian model, posing systemic risks to both individual rights and democratic institutions worldwide.
To address these threats, Lai proposed three policy recommendations for democratic governments: (1) banning Chinese AI products and services in government and critical infrastructure; (2) establishing an inbound review mechanism for Chinese AI services; and (3) building multilateral frameworks among like-minded countries to promote trustworthy cross-border data governance.
Participants raised questions on several topics, including censorship risks from Chinese open-source AI models, influence operations via large language models like DeepSeek, and Taiwan’s efforts to protect personal data.
In response to a question from participants about TikTok algorithm export controls, Huang noted the difficulty in verifying a potential transfer. He suggested that even if a sale occurred per U.S. requests, ByteDance might only transfer non-essential algorithms while retaining the core technology.
Another participant asked if Southeast Asian nations, which rely heavily on Chinese AI, could be incorporated into a democratic data protection framework. Huang responded that their deep integration into China’s data system would make adopting a U.S.-led framework difficult. A participant also added that U.S. AI companies may not be prioritizing Southeast Asian markets, contributing to the region’s reliance on Chinese services.
