While US AI companies navigate complex licensing deals with IP holders, Chinese firms like ByteDance appear to be using copyrighted material, such as specific actors' voices, without restriction. This lack of legal friction allows them to generate highly specific and realistic content that Western labs are hesitant to produce.
Related Insights
While generative AI introduces novel complexities, the fundamental conflict over artist compensation is not new. Historical examples, like musicians' families suing record labels over royalties, show these battles predate AI. AI's use of training data without permission has simply become the latest, most complex iteration of this long-standing issue.
Unlike Google and Meta who own vast video libraries, OpenAI lacked training data for Sora. Their solution was a legally aggressive "opt-out" policy for copyrighted material, effectively shifting the burden to IP holders and turning IP licensing, not just data access, into the next competitive frontier.
ByteDance's SeedDance 2.0 model integrates audio generation directly with video, a novel approach that suggests China may be starting to leapfrog the US in specific AI capabilities. This challenges the common narrative that China is only a fast follower in the AI race.
The perception of China's AI industry as a "fast follower" is outdated. Models like ByteDance's SeedDance 2.0 are not just catching up on quality but introducing technical breakthroughs—like simultaneous sound generation—that haven't yet appeared in Western models, signaling a shift to true innovation.
The geopolitical competition in AI will decide the economic value of intellectual property. If the U.S. approach, which respects copyright, prevails, IP retains value. If China's approach of training on all data without restriction dominates the global tech stack, the value of traditional copyright could be driven toward zero.
The market reality is that consumers and businesses prioritize the best-performing AI models, regardless of whether their training data was ethically sourced. This dynamic incentivizes labs to use all available data, including copyrighted works, and treat potential fines as a cost of doing business.
Leading Chinese AI models like Kimi appear to be primarily trained on the outputs of US models (a process called distillation) rather than being built from scratch. This suggests China's progress is constrained by its ability to scrape and fine-tune American APIs, indicating the U.S. still holds a significant architectural and innovation advantage in foundational AI.
The high quality of ByteDance's C-Dance video model suggests it may be trained on copyrighted material, like David Attenborough's voice, which US labs are legally restricted from using. This freedom from IP constraints could give Chinese firms a significant competitive advantage in media generation.
Companies like OpenAI knowingly use copyrighted material, calculating that the market cap gained from rapid growth will far exceed the eventual legal settlements. This strategy prioritizes building a dominant market position by breaking the law, viewing fines as a cost of doing business.
While an AI model itself may not be an infringement, its output could be. If you use AI-generated content for your business, you could face lawsuits from creators whose copyrighted material was used for training. The legal argument is that your output is a "derivative work" of their original, protected content.
