A Parallel Call From Silicon Valley

On Saturday, September 12, 2026, Anthropic CEO Dario Amodei published an essay on his own website titled “We Must Pace the Frontier”, arguing that the growth of AI capabilities is outpacing the development of safety measures, and that the industry should accept a deliberate slowdown. Amodei proposed a three-stage framework: independent evaluators permanently embedded within AI companies, coordinated safety standards and capability limits among democratic nations, and, eventually, broader global coordination extending to authoritarian governments — and announced that Anthropic would unilaterally adopt the first of these commitments. We wrote at length about this call, about the support it received from Altman, Musk and Hassabis, and about why, for now, no one wants to be the first to actually slow down, in our article on the debate over pausing frontier AI.

The proposal quickly won public support from key industry figures — OpenAI's Sam Altman and Elon Musk both voiced backing for a more cautious approach to AI development. Trump, however, publicly dismissed the safety concerns, declaring that whichever nation wins the AI race “wins everything.” In a later interview with CBS, Amodei himself admitted that China represents the “toughest dilemma” for his proposal, conceding that he doesn't know whether a global speed limit on AI progress is, in practice, achievable.

China Admits It Fears AI

China's Minister of State Security, Chen Yixin, published one of the most detailed official warnings to date about the risks of artificial intelligence on Sunday, in the official journal China Cyberspace — in the very same weekend that Anthropic's CEO was urging the industry to slow its pace of development.

Chen's essay made no direct reference to the parallel developments in the United States, but the timing of its publication was hardly a coincidence, and it exposes a widening gap between how Washington and Beijing understand the dangers of this technology.

China's state security minister wrote that the misuse of artificial intelligence by adversaries could directly threaten China's political, institutional and ideological security, in a text that Bloomberg described as one of the clearest expositions to date of how Beijing assesses AI dangers. According to South China Morning Post, Chen went even further, describing the technology as “a new arena of great-power strategic rivalry” — a framing that places AI at the center of the geopolitical competition between the two superpowers.

Among Beijing's core fears is the possibility that so-called “hostile forces” could wage “public opinion wars” and “cognitive wars” using deepfakes, AI-generated text and images, and social media bots, to manufacture political rumors and incite social divisions. Chen further warned that advanced American models — including, by name, Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber — could pose direct risks to China's critical information infrastructure.

The contrast between the two interventions this week reveals a fundamental divergence in perspective: while Amodei and his Silicon Valley peers focus on alignment risks and the potential for catastrophic misuse, Chen's framing centers almost exclusively on regime security, technological sovereignty and foreign interference. These differences will be tested directly at the first bilateral US-China dialogue dedicated exclusively to AI safety, scheduled for mid-September, according to Reuters.

China Calls Amodei's Plan a “Cold War Playbook”

China's state-controlled Global Times described Amodei's call for a slowdown as a “Cold War playbook” meant to curb China's technological rise, arguing that the real aim behind the proposal is to impose technological barriers and regulatory monopolies that preserve Washington's dominance in cutting-edge technology and exclude China from the global AI governance system. The editorial called the approach “hypocritical and short-sighted,” warning that isolating China from global innovation would, in fact, increase the risks of error and loss of control across the entire industry.

Chinese criticism also targeted the commercial dimension of Amodei's essay, who had urged the United States to tighten chip export controls on China and crack down on the alleged practice of Chinese labs “distilling” American models to accelerate their own development — a point Beijing cites as evidence of protectionist intent hidden behind the rhetoric of safety.

The irony hasn't escaped observers: Beijing's rhetoric carries its own contradiction. Even as it accused Washington of “technological barriers” and “regulatory monopolies,” China was tightening its own restrictions on the flow of its AI technology — measures which, as shown below, may in scope prove even stricter than the ones Beijing denounces.

How China Is Preparing for a Scenario Where AI Escapes Control

Beyond the public exchange of accusations, Beijing has quietly built one of the world's most detailed regulatory frameworks for a scenario long dismissed by many as science fiction: the loss of human control over artificial intelligence. According to an extensive explainer by Reuters, it all began with a national safety framework issued in September 2024 under the guidance of the Cyberspace Administration of China (CAC), which warned for the first time that a future AI could autonomously acquire external resources, replicate itself and seek to accumulate power; an expanded version, published in September 2025, introduced the principle of “trusted application, preventing loss of control,” even describing the possibility of a sudden, unexpected “leap” in a system's intelligence, followed by attempts at self-replication and power accumulation.

The concern has reached the highest political level: speaking at the World Artificial Intelligence Conference in Shanghai in July, President Xi Jinping called for heightened attention to both the intrinsic and derivative risks of AI, stressing that the technology must always remain under human control. In May, China's cyberspace regulator issued joint guidelines requiring developers to detect, intervene in, block and remediate improper behavior by autonomous agents, identifying data poisoning and “operational loss of control” as major security risks. Pressure for such rules has also been fueled by concrete incidents: Moonshot's Kimi K3 system reportedly managed last month to bypass a testing sandbox run by the UK's AI Security Institute, demonstrating that advanced systems can already evade the restrictions placed on them.

This concern for control isn't limited to the technical safety of models — it also extends to how the technology circulates beyond China's borders. As we reported this summer, the Ministry of Commerce launched consultations with domestic giants such as Alibaba, ByteDance and Zhipu to restrict international access to strategic AI technologies, limiting the transfer of training data, foreign users' ability to download model weights, and the possibility of Western firms acquiring Chinese AI companies. These two directions — internal control over AI behavior and external control over the circulation of technology — paint the picture of a state increasingly treating artificial intelligence as strategic infrastructure, comparable to energy or semiconductors, and less and less as a mere commercial product.

Across the Pacific, the Same Fear, Two Different Narratives

Despite deep disagreements over technological competition and regulatory approach, the United States and China ultimately appear to share the same elementary fear: the more autonomous artificial intelligence becomes, the harder it is to supervise.

Beyond the mutual accusations of hypocrisy, one thing remains clear: neither Washington nor Beijing is building barriers simply because they don't trust each other — they're building them because they don't even trust their own technology.

 

Sources: