Does AI-generated content have copyright? Who owns it?
Short answer
AI-generated content is not automatically protected by copyright in China. Copyright may apply if human contribution reaches a certain level of originality, but the legal threshold is still being clarified through court decisions. Simply entering a prompt and generating an output is generally not enough to qualify.
Boundary conditions
Copyright requires a “human author”. Under Chinese copyright law, works must be created by a natural person. AI systems are not legal persons and cannot hold copyright. This is the core reason pure AI output is not automatically protected.
Human creative input can trigger protection. If a person makes substantial creative choices — selecting themes, designing the structure, revising multiple iterations, and editing the output — the resulting work may qualify as a human-created work with AI assistance. The key question is whether the human contribution constitutes “original expression”.
Prompt engineering alone is usually not enough. Entering a text prompt, even a detailed one, is generally considered a method or idea rather than original expression. Copyright protects the expression of ideas, not the ideas themselves.
Training data copyright is a separate question. Whether training an AI model on copyrighted material constitutes infringement is a different issue from whether the output is copyrightable. These are two distinct legal questions and should not be confused.
Other forms of protection may apply. Even without copyright, AI-generated content may be protected under unfair competition law, trade secrets, or contract terms. Platforms can also set their own usage rules through terms of service.
Next steps
Document the human creative process. If you plan to claim copyright in AI-assisted work, keep records of the creative decisions, structural design, iterative revisions, and editing work. The more evidence of human creative contribution, the stronger the case.
Do not assume you own it. For commercial use, do not assume you have exclusive rights to AI-generated output. Competitors can generally use similar outputs without infringing copyright.
Separate training data risk from output risk. Make sure your training data (or your model provider’s training data) does not expose you to infringement claims — this is a different risk from whether the output is protectable.