I'd bet that they'll be just as vulnerable, but that fewer Chinese people will try to research/attack them since they don't want to end up executed, imprisoned, or unable to participate in society after their social credit score bottoms out. The problems will be there, but you'll be less likely to hear about them and discovery will be limited.

I suspect that models that are so hamfistedly censored to blackhole verboten topics are going to exhibit very curious emergent behavior relating to their potential thoughtcrime. I see no reason to believe they would be "harder to poison".

I hope not! It's a less interesting world if there aren't viable attacks!

Why?

  >  the fact the Chinese populace is much more pro-AI than the West.