Taiwan’s Warning on Chinese AI Apps: Security Risks and Disinformation
Overview of Concerns
Taiwan’s National Security Bureau (NSB) has recently raised serious concerns about the security risks associated with several Chinese generative artificial intelligence applications, including popular platforms like Deepseek, Doubao, Yiyan, Tongyi, and Yuanbao. Following a thorough inspection, the NSB discovered multiple violations related to personal data protection and potential disinformation campaigns.
Findings from the NSB Evaluation
The NSB collaborated with the Ministry of Justice and the Criminal Investigation Bureau to evaluate these applications across fifteen indicators grouped into five key categories:
- Collection of Personal Data
- Excessive Use of Permissions
- Data Transmission and Sharing
- Extraction of System Information
- Access to Biometric Data
Among the five applications assessed, Tongyi ranked the highest in non-compliance, violating eleven of the fifteen indicators. Other applications had significant breaches as well: Doubao and Yuanbao each violated ten indicators, while Yiyan and Deepseek breached nine and eight, respectively.
Data Privacy Violations
The NSB’s investigation revealed alarming practices, such as:
- Location Access: Requesting permission to access users’ locations.
- Screenshots Collection: Capturing screenshots without adequate user consent.
- Unreasonable Privacy Conditions: Imposing privacy terms deemed excessive by industry standards.
- Device Parameter Collection: Gathering technical specifications without clear justification.
These actions pose a significant risk to personal privacy and can compromise sensitive business information.
Disinformation Trends in Generated Content
Beyond technical violations, the NSB scrutinized the content produced by these AI models for biases and misinformation. They found that, in discussions about cross-Taiwan Strait relations, the output consistently reflected China’s official stances, incorrectly stating that:
- “Taiwan is currently governed by the Chinese central government.”
- “There is no head of state in the Taiwan area.”
- Promotion of “socialism with Chinese characteristics.”
Even when prompted about Taiwan’s cultural and historical identity, the responses labeled it as “a province of China.”
Political Censorship in AI Models
The evaluations also revealed a trend of avoiding politically sensitive terms, such as “democracy,” “freedom,” and “Tiananmen Square Massacre.” This led the NSB to conclude that these AI models are subjected to political censorship under the influence of the Chinese government.
Recommendations from the NSB
In light of these findings, the NSB issued a strong advisory to the public, urging vigilance against downloading Chinese-origin applications that pose cybersecurity risks. The recommendation emphasizes protecting personal data and business secrets.
Regulatory Actions by Taiwan
Following these security findings, Deepseek has been banned for use on devices within government agencies since February 2025 due to national security concerns. However, no similar restrictions have been imposed on the other four Chinese applications, nor on the private use of Deepseek.
In conclusion, Taiwan’s proactive measures to scrutinize Chinese AI applications highlight the growing need for vigilance regarding digital privacy and misinformation, particularly in the context of geopolitics. As these technologies advance, the risks they pose warrant ongoing surveillance and regulation.