China's AI Safety Debate: What It Means for You

The author walks through a systematic look at the status quo of Chinese attitudes toward AI safety and development — while calling the piece's own title inflammatory and clickbaity. Among the highlights: Zhipu is the only startup of those covered that is fully entity-blacklisted by the US, and the author is impressed by its cyberdefense work ahead of publicly releasing GLM-5.3, as well as its safety-pilled attitudes and commitments compared with other Chinese AI companies.

Even so, Anthropic took a jab at Zhipu rather than aiming it at less safe companies — criticisms the author says have some merit and deserve serious consideration, and which a follow-up post will address with the current evidence and arguments for and against open-weight AIs in terms of existential risks. For AI experts in China, the author's message is that now is probably the time to do something: join safety-pilled Chinese orgs like Concordia AI, independently publish papers, apply for grants, and connect with like-minded Chinese researchers who are particularly worried about existential risks.