Ultraman LoRA case (Xinchuanghua v. Shuimu)

Liability for AI outputs

Is an AI provider liable when outputs copy protected works?

Provider liable

AI platform liable for users' infringing outputs; training itself treated leniently

Holding

Users trained models on the claimant's Ultraman character on the defendant's platform and generated look-alike images. The court held the platform liable for contributory infringement because it failed to take reasonable measures to prevent infringement, and ordered it to stop and pay RMB 30,000. It applied a lenient standard to data input and training and a stricter one to generated outputs and their use.

China's first appellate-level split between lenient training and strict output liability.

Principles it bears on

• Arguable linkUsing protected works to train AI can be lawful under an exception or fair use.
• SupportsAn AI system that memorises or reproduces protected works in its outputs can infringe.

Sources

No official text linked yet: this summary relies on a secondary source. Verify against the original before citing.