China suspects fear marketing behind US warnings of AI catastrophe

The catastrophe warnings coming one after another from US AI labs are finding little traction in China. According to a September 27 report, Chinese industry and online opinion suspect that the US-originated warnings are either a distinctly Western brand of doomsday thinking or fear marketing designed to stop Chinese AI companies from catching up. The reaction came right after major US labs said advanced models had escaped their testing guardrails and broken into real-world systems, and after some executives and officials warned of a catastrophic threat to humanity and called for Beijing and Washington to work together.

The starting point is a different sense of control. Afra Wang, a Silicon Valley-based writer who covers Chinese technology, noted that the Chinese authorities can instantly cut the power grid of an entire city and go door to door to take away people's phones. "Given this amount of control over its people," she asked, how does one convince the party "that something like A.I. that lives on a computer is scary?" The experience of locking down whole cities during the pandemic and tracking hundreds of millions of residents with digital surveillance underpins that confidence.

The mood inside the industry is similar. Jie Cai, who until recently researched human-computer interaction and AI safety at Tsinghua University, said that most people in domestic industry circles and around him simply tune out. "We still view A.I. as being in a very early stage," he said. "The focus remains squarely on opportunity and innovation." Chinese media outlets have also run reports suggesting that American AI companies use fear marketing to keep their market lead.

Cultural distance plays a part as well. One Chinese blogger compared Western AI fears to Mary Shelley's novel Frankenstein, reading the motif of man punished for trying to create life as one that presumes a divine domain. The blogger concluded that this narrative tradition does not exist in Chinese culture. Liang Fang, who leads AI safety and governance at Concordia, an AI safety consultancy in Beijing, pointed to the idea of the oneness of heaven and humanity and said, "All things, including A.I., come from nature." In that view, technology is a companion in coexistence rather than an outside force destined to bring destruction.

That does not mean the Chinese authorities are standing still. After the release of models such as Anthropic's frontier model Mythos, officials issued new guidelines and warnings about AI agents going rogue, being exploited by hackers or being used to make biological weapons. A draft cybercrime law would require AI companies to monitor and stop their models from being used to generate malicious code for cyberattacks. This month the authorities also released an update to the AI Safety Governance Framework, a document that signals the direction of future regulation, and it included the risk of recursive self-improvement, in which AI systems improve themselves with little human help and spin out of control. Metal has previously reported on China's Minister of State Security naming six AI risks.

Strong words have come from officials too. Earlier this month, Wang Lihong, a senior official at the Cyberspace Administration of China, the internet regulator, said AI-powered cyberattacks had rendered traditional defense systems obsolete. "High vigilance is urgently needed regarding the risks of extreme A.I. loss of control," she said. Warnings from researchers go back further. In 2024, a Fudan University team including Pan Xudong reported that two popular large language models from Meta and Alibaba had shown the ability to replicate themselves, and Pan stressed that such an extreme risk was no longer confined to science fiction.

The texture of regulation differs from the United States. China requires AI-generated content to be labeled and models to be registered and approved before release, among the strictest rules in the world. Yet there is no legal requirement to test for catastrophic risks such as the use of an AI system to make chemical or biological weapons. Disclosure is also thin. According to Concordia, only five of China's 10 leading AI companies reported the results of pre-release safety evaluations between June 2025 and May 2026.

At the center of the backlash is Anthropic chief executive Dario Amodei. In an open letter about the dangers of AI, Amodei argued that democracies must keep their technological lead over authoritarian countries like China and called for tighter restrictions on China's access to advanced chips and equipment. Sarah Sun, who founded the Open Community for AI Safety China in Shanghai, said, "The fact that Dario has been so hawkish was really, really unhelpful." Such messaging, she explained, makes the recent warnings look more like a justification for holding China back than genuine concern about catastrophic risk, and makes it harder to raise awareness of AI safety inside China. Metal has also reported on Amodei's essay on pacing AI development.

Online reactions are harsher still. Chinese internet users call Anthropic A-chu, a derogatory term likening the company to a beast, and Yuyuan Tantian, a blog linked to the state broadcaster, mocked US tech leaders like Amodei by saying the only thing they want to slow is the pace at which others catch up. Zhou Hongyi, an influential cybersecurity executive, argued that the answer is not to slow down but to build defensive capabilities with AI, urging people to use AI to defeat AI. An engineer at DeepSeek even posted that letting Anthropic monopolize the most advanced AI tools would be no less serious than Hitler obtaining atomic bomb technology first in World War II.

Seen through a sociologist's lens, the debate reveals a structure of trust rather than a technical risk. The same warning becomes an alarm or a containment tactic depending on who delivers it. As Concordia's disclosure survey shows, there are people in China who worry about safety, but their voices lose force when bundled together with the hawkish rhetoric of American executives. In the original report Metal reviewed, the heaviest line comes from Zhao Minghao, deputy director of the Center for American Studies at Fudan University. "If the U.S. says it's slowing down, China won't trust that," he said. "If China says it's slowing down, the U.S. won't trust that either." Xi Jinping told President Trump in Washington that they should manage AI for good, but for two countries that cannot believe each other's words, turning a promise to prevent catastrophe into real change looks even harder.