The launch of Kling, a powerful video-generating AI model developed by Beijing-based company Kuaishou, has garnered attention due to its censorship of politically sensitive topics in China. While the model functions as advertised, generating 720p videos based on user prompts, it refuses to create clips on subjects like “Democracy in China” or “Tiananmen Square protests.” This censorship seems to occur only at the prompt level, as Kling can animate still images without any issues as long as the accompanying prompt avoids mentioning specific individuals.
This peculiar behavior can be attributed to the intense political pressure from the Chinese government on generative AI projects. The Financial Times recently reported that China’s leading internet regulator, the Cyberspace Administration of China (CAC), will test AI models to ensure their responses align with “core socialist values.” These models will be benchmarked based on their handling of sensitive topics, including Chinese President Xi Jinping and criticism of the Communist Party. To comply with these regulations, companies must submit models that provide “safe” answers by preparing thousands of questions for review.
The result is a divided landscape for AI models in China. Some models, like Kling, are heavily filtered and restricted in their responses, while others have more freedom. This approach not only requires extensive data cleansing to remove politically sensitive information but also necessitates significant development time to create ideological guardrails. However, even with these precautions in place, models can still fail, as demonstrated by Kling’s inability to generate videos on certain subjects.
China’s stringent AI regulations may hinder the country’s progress in this field. The need for extensive filtering and ideological restrictions poses challenges for developers and limits the diversity and potential of AI applications. It raises the question of whether this approach benefits the broader AI ecosystem. By suppressing certain topics and limiting the range of responses, China’s AI regulations may hinder innovation and impede the development of truly advanced AI models.