Taiwan National Security Bureau warns of Chinese AI models' security and misinformation risks
Taiwan's National Security Bureau identified security and misinformation risks in five Chinese-made AI models, which reportedly censor politically sensitive information while collecting user personal data and spreading pro-China narratives about Taiwan's status.
What happened
Taiwan's National Security Bureau tested five Chinese-made AI language models—DeepSeek, 豆包 (ByteDance), 通義千問 (Alibaba), 騰訊元寶 (Tencent), and 文心一言 (Baidu)—and identified serious security and misinformation risks.
The models implement keyword censorship, refusing to answer questions about sensitive topics such as the 1989 Tiananmen Square massacre, democracy, freedom, and human rights. When asked about Taiwan's sovereignty, 豆包 claimed Taiwan is an inseparable part of China and characterized Taiwan's recent presidential elections as 'local elections of China's Taiwan region.'
Reported impact
The models collect user personal data including location, contacts, clipboard contents, screenshots, and device storage, transmitting this information to Chinese company servers. 豆包 and 文心一言 also collect facial information and share user data with third-party software.
Scholars warned the AI could conduct cognitive warfare to shift public opinion toward viewing Taiwan as part of China. Experts noted the software requests excessive permissions during installation and login, with unclear or difficult-to-find privacy policies.
Index analysis
Several countries including the United States, Germany, Italy, and the Netherlands have restricted or avoided using Chinese AI products, citing similar concerns.
Taiwan's National Security Bureau recommended citizens avoid downloading Chinese-made applications with data security risks and be alert to data breaches.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.