Skip to content

Product1 publisher2 min readPublished

Beijing aims its export-control script at Anthropic's chief executive

Dario Amodei published a chip-export ask on Saturday. By Monday, China's commerce ministry was answering it in the language it normally reserves for Washington. State media gathered three researchers to make the case.

The Product Desk · Product desk

Illustration accompanying Beijing aims its export-control script at Anthropic's chief executive

What happened

  • Dario Amodei's Saturday essay asked Washington not to sell powerful AI chips or semiconductor manufacturing equipment to China, and to crack down on smuggling and on remote access to data centres abroad.
  • China's commerce ministry called the essay "further proof that the US is pursuing technological hegemony in AI", in remarks the Global Times carried on Monday.
  • The paper built a full report around the theme, gathering three Chinese researchers to argue that the Anthropic chief is asking for cooperation and containment at the same time.
  • A separate Global Times commentary circulated on WeChat called the plan a "quiet AI Cold War" and described it as hypocritical and short-sighted.
  • Beijing has contested export controls as government policy for months, and this time the ministry pointed that standing script at a private chief executive over a post he published himself.

Compiled by The Product DeskSomething wrong?How this is made

Why it matters

  • exposure A US lab's export-control asks are now an attribute of its product for anyone distributing near China, because the safety commitments and the chip denial arrived in one document.
  • contradiction Trump dismissed the slowdown case last week because America has to win. Beijing read the same essay as a plan for America to win. Amodei is being answered from both directions at once.
  • constraint The one step Amodei has not seen matched inside the industry needs the participation of the government that has just called his essay hegemony, so coordinated pacing starts from a public rejection.
  • decision Teams planning compute supply or China-facing distribution have to treat vendor policy essays as scheduled inputs, since a Saturday post by one chief executive set Monday's line from a foreign trade ministry.

Xiao Qian of Tsinghua's Institute for AI International Governance made the point a procurement team should keep hold of: Anthropic competes with Chinese open-source models [7]. That does not make the safety case insincere, but the party asking for limits on Chinese chips, computing power and models competes with the models it wants limited.

Xiang Ligang, a telecoms and technology policy observer, called the proposal groundless and hostile [5]. The charge underneath the label is more useful than the label. Xiang said Amodei accepts that pacing AI would eventually require cooperation with China while advocating restrictions on chips, computing power and models to slow China down, and openly seeking to widen the American lead [6].

The version with the longest reach came from Liu Shaoshan, who directs embodied AI at the Shenzhen Institute. When AI is cast as the technology that settles national security and the future of civilisation, he said, the company building it stops being an ordinary tech company and becomes "strategic infrastructure" [8].

None of the underlying position is new. On 9 September the NSA, FBI and CISA named six Chinese firms over industrial-scale distillation, and Beijing called the accusation unfounded the next day [12]. In June, Wang Yi pitched a global AI cooperation organisation offering cheap models to developing countries while the G7 argued over who should get access to American ones [13].

A Public First survey of 18,000 people across 15 countries, published days after that, found most already believed China had pulled ahead on capability [14]. That averages about 1,200 respondents a country [17]. The survey measures belief. Buyers carry that belief into the meeting where they weigh a US vendor against a domestic open-source model.

Third-party evaluators with employee-like access is a commitment Anthropic has made on its own [10], which means a customer can raise it at renewal and expect an answer with a date attached. Common safety standards among frontier labs in democracies moved inside the industry within a day, with Sam Altman matching the first two steps the same Saturday [11]. Chip sales and pacing sit with governments, and a buyer can only budget for those.

What to watch

  • Whether the commerce ministry follows the remarks with an actual measure touching US model providers, such as procurement or licensing restrictions.
  • Whether Anthropic publishes terms for the third-party evaluator access it has committed to, in a form a customer can cite at renewal.
  • Whether other Chinese agencies start answering individual lab executives by name, or keep their responses pointed at Washington.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories