US officials accused six Chinese AI companies—including DeepSeek, Moonshot AI and Alibaba—of using outputs from American models for unauthorized industrial-scale distillation. Distillation itself is a standard technique for transferring behavior into a smaller model; the dispute concerns authorization, intellectual property and alleged government involvement.
What changed
Model-output provenance is moving from a provider-policy issue into international enforcement and security policy.
Why it matters
API outputs can function as valuable training data even when weights, code and original datasets remain private.
Practical takeaway
Record whether generated data may be used for training, its originating model and licence, and the consent basis; make those fields mandatory in every synthetic-data manifest.
Analysis
The accusations have not been independently proven, and technical similarity alone does not establish unlawful copying.
Published
8 September 2026
