ChinaTalk Author Irene Zhang: Why Chinese AI Researchers Read US Safety Warnings as an Anti-China Conspiracy

On September 13, 2026, the day after the release of DeepSeek V4.1-Flash, kernel researcher Liu Shengyu published a blog post on his personal WeChat Official Account, which quickly spread through China's AI circle. He predicted that "in another six months or a year, AI-written kernels will most likely be just as good as mine, if not surpass my work," and then wrote: "If Anthropic forever controls the world's most advanced AI, care to guess whether future society would turn out communist or Cyberpunk 2077?" His answer was the latter, stating that "the severity of that would be no less than letting Hitler get the atomic bomb before the Allies." His reason for staying at DeepSeek was: "We research powerful, fast, universally accessible AI and open-source it, and perhaps that can pull the world back a little from the 2077 end."

This blog post received overwhelming support on Zhihu, Xiaohongshu, and in blog comment sections. People said that DeepSeek demonstrated the greatest commitment to an "inclusive vision," that Dario Amodei's risk warnings were "crying wolf," that OpenAI and Anthropic were "burning the zither and stewing the crane" (wantonly destroying all that is beautiful), and that Anthropic, in particular, was labeled "techno-fascist" and "barely human." ChinaTalk author Irene Zhang, in an analysis article published on September 18, pointed out that this group is precisely the one with the closest contact with the US AI community, arguably possessing the most humanizing and nuanced perspective, yet they are concluding that the other side is driven solely by anti-China ideology.

Zhang's core argument is not simply that "China has conspiracy theories," but rather identifying a specific interpretive mechanism. She recalled overhearing a session at the ICML conference in Seoul earlier that summer, where a Chinese practitioner was completely baffled after listening to her explanation of Amodei's worldview: "It's that simple? Then... nothing else?" Amodei's logic, in Zhang's view, is indeed simple—liberal democracy plus AGI equals export controls against China—a deduction driven by sincere belief, rooted in effective altruism and East Coast liberal internationalist traditions. However, Chinese analysts are accustomed to interpreting public statements through the rhetorical framework of "in the name of...," tracing back from Mao Zedong's 1939 criticism of the US "in the name of neutrality" to Deng Xiaoping's 1974 criticism of the Soviet Union as "socialism in words and imperialism in deeds." This tradition treats public declarations as cover, with true motives hidden behind the scenes.

Zhang points to a key turning point: previously, China's conspiracy narratives relied on tenuous connections—between Western journalists and their governments, between Hong Kong democrats and foreign organizations, and between European companies and the EU. But this time, Washington and Silicon Valley have publicized their relationship from the start. Amodei posted six articles on his personal blog, each advocating for export controls to China; most of the national security community's evidence regarding Chinese distillation was provided by Anthropic itself, rather than resulting from FBI investigations or Congressional inquiries. While US AI labs may view these disclosures as progressive moves toward accountability by a new generation of industry leaders, in China's eyes, they are evidence of coordinated action between Washington and Silicon Valley.

More alarming to Zhang is that young researchers are reaching conclusions even more radical than those of the Party itself. Liu Shengyu's blog post, calls to boycott Anthropic, and mocking interpretations of safety warnings all indicate that the youngest generation of Chinese AI researchers is turning the Party's conspiracy narrative into their own. They are no longer just soldiers of state projects, but seem to believe that a race to the bottom is inevitable, and have internalized Western discussions of authoritarian AI threats as a coordinated attack on their right to technological advancement. Zhang wrote: "The Party's conspiracy is now the technologist's own conspiracy." She believes that anyone who believes Chinese participants are essential to navigating the transition to a world with much more powerful AI systems should be troubled by this.

Zhang does not push the conclusion to pessimism. She notes that senior official Wang Lihong of the Cyberspace Administration of China listed AI governance challenges in a Concordia AI briefing—inherent unreliability, misuse in biology and cyber, and increasing autonomy—which almost perfectly correspond to expressions in Amodei's own blog post, "The Urgency of Interpretability." "Someone at the CAC is reading his essays; apart from export controls, they also want to win in the safety field." She believes that acknowledging the competitive nature of bilateral relations might actually make useful actions easier to advance. But she leaves an unresolved question: "Races aren't always aimed at the bottom. What would a truly grounded strategy to engage China look like?"

The strongest rebuttal to this analysis lies in the fact that Zhang's arguments rely mainly on one viral blog post and several online comments, and "Chinese AI researchers" are not a monolith. She herself acknowledges the existence of "a minority of Chinese commentators sympathetic to Coxon" and "more sober professional tech media" that accept the premise that AI could bring unprecedented harms. Generalizing Chinese analytical tradition as "conspiracy-theory-style" may itself constitute another simplification. Furthermore, it cannot be entirely excluded that some of the Chinese side's suspicion of US motives does have factual basis—when a company simultaneously advocates for export controls, self-reports safety findings, and competes commercially, the boundary between "transparency" and "strategy" is indeed blurred.