Chinese state media and industry voices now describe U.S. calls to slow frontier AI as cover for tighter chip export controls. The New York Times ran the story on September 27. The piece recasts public safety arguments from American labs as a deliberate effort to preserve Washington's lead in advanced compute hardware.

What does Anthropic want?

The clearest evidence sits in Anthropic's dual track. CEO Dario Amodei has publicly backed joint safety standards and speed limits on frontier systems. He has also pushed, repeatedly, for stronger export controls on advanced AI chips and on the equipment used to build them. Beijing treats those two positions as parts of one strategy, not separate concerns.

Takeaway: a safety advocate and an export-control hawk can be the same person speaking from one podium.

How does Beijing answer?

Chinese industry has answered with the opposite call. Zhou Hongyi, founder of cybersecurity firm 360 Group, argued recently that "we must defeat AI with AI." He pressed for defensive AI systems to counter offensive ones. Faster domestic development, not a pause, is the answer his camp favors.

Takeaway: in Beijing's reading, slowing AI is a Western privilege Beijing will not grant itself.

Where does the perception gap sit?

It sits on what AI risk actually means. U.S. labs describe existential risk from runaway systems, scenarios in which models outgrow human oversight. Chinese commentators point to government control over the internet and over private firms as proof that runaway AI is manageable. The Times quoted one passage saying the Chinese government has authority "to the extent of cutting off a city's power grid and seizing individuals' mobile phones when necessary."

Takeaway: one side worries about an ungovernable machine, the other about a machine under different governors.

Why does this matter for chips?

Because the AI safety debate now moves inside export-control policy. Anthropic and peers are calling for slower compute growth at home while seeking to keep advanced accelerators out of Chinese hands. The policy covers both finished AI chips and the production equipment used to build them. Every safety statement lands inside a trade argument over who can buy the silicon and the tools that build it.

Takeaway: every safety hearing in Washington is now also a hearing on chip shipments to Chinese fabs.

Can joint standards happen?

Not soon, on current evidence. The Center for a New American Security has stated that, without mutual trust, agreements on joint AI safety standards or development speed control are hard to reach. That diagnosis closes the loop. The same distrust that blocks standards also blocks any pause in chip restrictions, which in turn feeds the Chinese belief that U.S. safety talk is strategic.

Takeaway: the two debates are linked, and so are their obstacles.

The next test will come when Anthropic or its peers file the next round of export-control comments, and when Beijing answers in kind at the United Nations, in trade filings, or in chip-equipment licensing decisions.