LessWrong Author CMLKevin Explains Structural Headwinds in Chinese AI Safety Field
CMLKevin highlights that China's frontier AI safety field faces significant structural barriers, including funding scarcity, restricted model access, and risks associated with foreign outreach. Many researchers are forced to fund non-profit safety work through commercial products, a dynamic often misunderstood by the West as purely capabilities-driven or state-aligned. The post calls for supervised compute and subscription support for individual researchers.
ImportanceMaterialEvidenceE3 inspectableWrite-upQuick
China is weak in frontier AI safety, and the domestic field is currently facing significant challenges and headwinds. After talking with various researchers on the Chinese side, LessWrong user CMLKevin summarizes the situation. He notes that Chinese AI startups and hyperscaler labs have very little frontier risk situational awareness. This lack is generally not due to deliberate ignorance or strategic avoidance, but rather a "complete vacuum of imagination." In contrast, the Western field had FOOM debates in 2011, while China had nothing analogous in mainstream AI research consciousness before 2023.
Getting any funding at all is hard. Being a non-profit means a lens of suspicion by default. Getting foreign EA funding/grants is basically a non-starter given China's stringent foreign non-profit rules. Domestically, there is very little local EA money. Therefore, many safety-pilled people are forced to do for-profits, selling safety & observability products or courses to domestic lab employees, while using most profits to conduct basically non-profit safety research. CMLKevin argues this is an underdiscussed aspect of the Chinese ecosystem, and fears the Western sphere may misunderstand the phenomenon as "this is just ultimately capabilities research or working for the state's goals."
Getting frontier AI assistance is difficult and costly. There are no "Claude for Chinese safety researchers" carve-outs; in fact, accounts are regularly banned even for those conducting clear civilian AI safety work, simply because the platform detects the user might be from China. Since researchers still need the best models to assist with research, this measure incurs disproportionately high costs on those with less funding, discouraging them early.