In September 2025 the Supreme People’s Court released a set of typical unfair competition cases. One judgment involving an AI model drew attention. The facts were simple: one party built a cartoon-effect model through data training and parameter tuning, and the other directly used its model structure and parameters to launch a competing product. The court held that the model formed through that investment was a core resource conferring competitive advantage, and therefore a protected competitive interest, and that the direct use constituted unfair competition.
Why the holding matters
Protection of AI models has long been unclear. Copyright protects the expression in code, but a model’s value often lies in the parameters and structure produced by training, not in specific lines of code. Patent protection requires novelty and inventiveness that training methods or architectures may not satisfy. Trade secret protection requires secrecy and confidentiality measures.
The significance of this case is that the court did not force the facts into a single right. It reasoned from “competitive interest”: where a party has created something through lawful investment that confers competitive advantage, that result can be protected under the Anti-Unfair Competition Law even if it does not fit neatly within copyright or patent. See our note on IP protection for AI companies.
Three pillars of the finding
Substantial investment. The model’s value comes from sustained training and tuning, not from simple functional imitation. That is exactly what the court treated as a differentiated competitive resource.
Direct use. The defendant did not develop independently but used the model structure and parameters directly — a “take and use” that undermined competitive order.
Competitive harm. Direct use let the defendant skip the development investment and gain a market advantage — classic free-riding.
Three takeaways for AI companies
First, put model assets on the protection list. Beyond code and documentation, model structure, parameters, training data recipes and tuning records should sit within internal asset management and confidentiality. See our IP protection service.
Second, retain evidence of the development process. Data sources, training logs, parameter versions and tuning records are the factual basis for asserting a competitive interest. Without process evidence, “substantial investment” is hard to prove.
Third, be clear-eyed about open source and collaboration boundaries. Using third-party open source models requires strict compliance with licence terms, and collaboration agreements should specify model ownership and downstream use. See our note on AI commercial contract clauses.
Together with earlier AI copyright cases, this ruling sketches a trend: judicial protection in AI is moving from a single right type towards a combined assessment of investment, result and competitive order. See our notes on the first AI model infringement case and the Beijing court’s AI copyright criminal case.
This article is based on the typical cases released by the Supreme People’s Court and on public information. It is for general reference only and does not constitute legal advice.