Skip to content

Translation in progress Some entries still include text in the original language.

安远AI· Gabriel Wagner·Indexed by this site · Originally published

中国开源最强 AI 模型,安全吗?

China is open-sourcing its best AI models. Is that safe?

AI summary

中国实验室今年夏天发布 Kimi K3、Qwen 3.8-Max、GLM-5.3 等开源模型,能力仅略落后于美国最强模型,引发开源模型是否会被滥用的争论。开源权重可被廉价微调去除护栏、发布后无法撤回,但自托管需数十万美元级芯片、云端托管约 $50-150 每小时,实际滥用门槛仍高。中国 TC260 已在 AI 风险框架中点名开源模型安全机制可能被移除或绕过,并着手起草开源 AI 安全标准。

Related papers
Read originalaisafetychina.substack.com