I strongly believe that's the case. That's even the case with the recent OpenAI hack of HuggingFace.

The censorship of US models is so bad (read: guardrails), that they had to use GLM 5.2 on their own hardware to do the analysis. That's how locked down, restrictive, and closed source American models are. They're basically for-profit piracy by way of selling tokens.

Whereas the Chinese models are just "here you go, download and run".

I know what I run. Qwen3.5 and 3.6 and GLM5.2 . USA token vendors are incentivised in doing worse to sell more slot machine tokens.