Dario Amodei used a weekend essay to ask the world to slow AI development, and used the same essay to ask Washington to hold China back. By Monday Beijing had its answer ready. Foreign ministry spokesperson Guo Jiakun told a press conference that intimidation, confrontation and fierce competition would only disrupt global AI governance and serve no country's interests. The harder reply came from the security apparatus, where the head of the Ministry of State Security used a state journal to name Anthropic's Claude and OpenAI's GPT-5.5-Cyber as weapons pointed at China.
Source: theguardian.com
The Anthropic chief executive's argument runs in two directions at once. He warns that a swarm of AI agents could take over the internet within six to twelve months unless researchers agree among themselves to slow capability development. He also warns that AI could hand the Chinese Communist Party the ability to "militarily dominate democracies," and that Chinese leadership in AI would be a serious threat to the United States and the world. He concedes that any coordinated slowdown would require cooperating with China, and in the same essay calls for preserving the democracies' lead over the autocracies as far as possible.
Other technology executives lined up behind the essay in general terms while declining to touch the China passages — Sam Altman of OpenAI, Google DeepMind chairman Demis Hassabis, and Elon Musk of SpaceX among them. The selective endorsement is the tell. Slowing down is a position a lab can hold in public at no cost. Saying out loud that a specific foreign government must be prevented from acquiring the same capability is a geopolitical commitment, and none of them made it.
Beijing's official line separates the two halves. It rejects the claim that Chinese dominance in AI would be dangerous, while sharing the worry about how fast the technology is moving. At a conference in July, Xi Jinping said AI must always remain under human control.
What the Chinese government is actually afraid of surfaced in an article published Sunday in the state journal China Cyberspace. Chen Yixin, head of the Ministry of State Security, wrote that the technology could threaten the government's grip on power. He said "hostile forces and people with ulterior motives" are using AI for propaganda and cognitive warfare against China, and that this directly threatens the country's political, institutional and ideological security. He singled out two American models, Anthropic's Claude and OpenAI's GPT-5.5-Cyber, claiming they had been turned into weapons and had become examples of a new class of cybersecurity threat.
Read side by side, the two documents are not two positions in one debate. They are two debates wearing the same word. Amodei's risk is that capability outruns control and nobody can stop it. Chen's risk is that capability reaches Chinese citizens and the Party cannot filter it. A slowdown agreement built on the assumption that both sides want the same thing has nothing underneath it, because the thing Beijing wants controlled is not the thing Amodei wants controlled. Notably absent from the essay is any account of what China would be agreeing to, or why a state that has just catalogued Claude as a weapon would coordinate with the company that makes it.
In Washington the response was to treat the whole category as noise. On Monday Donald Trump said AI needs nothing more than a "strong and smart" president as a check on it. That was the same week shares in AI-linked companies fell on concerns about the pace of the industry's growth — a market Trump watches closely, which makes his dismissal cheaper to read as reassurance than as policy.
China's state press went further than the foreign ministry. In an editorial published Sunday, Global Times called the essay a veiled call for containment, saying the text was packed with anti-China provisions and amounted to a "Cold War script" for the AI industry.
The leaders of the two countries are expected to discuss AI governance when they meet on September 24. They will do so a week after the leading American safety advocate asked for a deal with a country he also asked to be contained, and after the Chinese state security apparatus put his company's model on a list of hostile instruments. Anything agreed on the 24th has to be built on top of both of those documents.