Taiwan Pushes Meta and Google to Label AI Images to Curb Deepfakes

The rapid advancement of artificial intelligence in recent years has propelled generative AI to the forefront of social media. From voice synthesis and text generation to deepfakes of images and videos, these technologies not only simulate human creation with high fidelity but also pose significant risks to information security and social trust. In particular, the emergence of deepfake videos has been demonstrated to spread misinformation rapidly, posing potential threats to public opinion, electoral integrity, and personal privacy. In response to these challenges, governments worldwide have begun reviewing their digital governance frameworks and implementing stricter content moderation and labeling regulations for social platforms.

In Taiwan, the National Development Commission (NDC) announced a three-tier mechanism in early this month to compel platforms such as Meta and Google to implement labeling for AI-generated images, aiming to enhance the transparency of AI-generated content to a level where users can identify it in real time. The NDC noted that with the widespread application of generative AI in commercial and entertainment sectors, the failure of platforms to clearly label AI-generated images could lead to information confusion and misdirection. Therefore, the NDC will adopt a dual approach of policy promotion and technical collaboration to ensure that platforms complete labeling before content distribution and provide users with convenient mechanisms for reporting and filing complaints.

Specifically, the three-tier mechanism consists of the following: The first tier requires platforms to provide a "user-initiated labeling" feature, allowing content creators to manually add tags such as "AI-generated" or "Deepfake" when uploading AI-generated images, along with displaying the generation model and parameters. The second tier involves "technical detection" to automatically identify AI-generated images and automatically add corresponding tags on the platform interface, reducing the burden on users to label content manually. The third tier establishes "reporting and complaint channels," enabling users to directly file complaints with the platform or the NDC when they discover missing labels or suspected violations, with dedicated teams conducting reviews and processing. These three tiers complement each other, jointly building the traceability of AI content from the perspectives of creators, platforms, and users.

At the technical level, AI image detection still faces issues of high error rates and false positives. Although multiple studies have developed detection models based on frequency domain analysis, optical features, and deep learning, practical applications still require fine-tuning and optimization for different generation models, such as Stable Diffusion and Midjourney. The NDC stated that it will collaborate with academic institutions domestically and internationally to establish open datasets and evaluation standards to promote the continuous iteration of detection algorithms. Additionally, when platforms provide the "user-initiated labeling" feature, they must consider user experience and privacy protection to avoid reducing creators' willingness to upload content due to overly cumbersome labeling mechanisms.

For international platforms such as Meta and Google, this measure is not only a compliance requirement but also a key aspect of corporate social responsibility and brand image. Timely implementation of labeling mechanisms will help enhance user trust in platform content and reduce legal risks associated with fabricated content. On the other hand, as Taiwan is positioned as a "policy hub for AI industry innovation," leading the way in implementing AI-generated image labeling will provide a safer and more transparent creative environment for local entrepreneurs, further attracting international investment and technical talent.

In summary, the NDC's three-tier mechanism is not only a governance measure for AI-generated images but also an enhancement of the transparency and credibility of the overall digital ecosystem. Through the collaboration of creators, platforms, and users, Taiwan is expected to set an advanced benchmark in global digital governance, laying a solid foundation for broader future regulations on AI content. (Source: Central News Agency)