Anthropic’s Dario Amodei responds: doesn’t oppose open-weight models, but fears Chinese AI
Anthropic CEO Dario Amodei has publicly denied that his company supports banning open-weight AI models, following industry speculation prompted by an open letter from Nvidia, Meta, Microsoft, Hugging Face and Mistral urging policymakers against "premature restrictions" on such models. In a blog post published on Monday, Amodei clarified that Anthropic has never advocated for such a ban and that he views open-weight models without dangerous capabilities as a public good, distinct from the threat he believes authoritarian states pose.
Amodei's real concern lies with "authoritarian governments" — particularly the Chinese Communist Party, which he called the "most capable" such actor — potentially using advanced AI to secure military dominance or repress their populations, including through biological or cyberattacks. He argued open-weight models are harder to safeguard once released, citing a UK AI Security Institute report noting they cannot be withdrawn, and called for restricting China's access to powerful chips and cracking down on "distillation" techniques used to copy rival models. He also voiced support for a global AI safety-testing body, provided China participated too, suggesting cooperation on biological weapons risks could be achievable.
In short
- Amodei denies Anthropic ever backed banning open-weight AI models.
- His real worry is authoritarian states, especially China, misusing AI.
- He backs curbing China's chip access and a global AI safety-testing body.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Advocates of open-weight AI argue that publishing model weights broadens access to powerful tools, lets independent researchers scrutinise and improve safety, and prevents a handful of large labs from monopolising a transformative technology. They contend that openness accelerates scientific progress and economic opportunity worldwide, and that since capable open models already circulate globally, restricting them mainly disadvantages legitimate developers while doing little to slow determined state actors.
The case against
Others, echoing concerns Amodei has voiced, argue that releasing frontier-level weights risks handing adversarial states, particularly China, capabilities that could erode the West's strategic, military and economic edge, since safeguards built into a model can often be stripped out once its weights are public. They believe that in a competitive geopolitical environment a lead in advanced AI is a genuine national asset, and that caution about what is released, and when, is a prudent trade-off even if it slows the pace of open diffusion.
Explained simply
Open-weight AI models are artificial intelligence systems whose underlying code is publicly released and available to download. There is growing debate about whether governments should restrict the development and distribution of these models, with the technology industry divided on the question.
Several major technology companies including Nvidia, Meta, and Microsoft recently urged policymakers against imposing restrictions on open-weight models. Anthropic, an AI safety company, is among other significant players in ongoing discussions about how artificial intelligence should be governed.
Supporters of open models emphasise benefits for innovation and competition, while critics worry about potential misuse by hostile governments seeking military advantage or surveillance capabilities. These concerns about which countries might gain access to powerful AI technology are at the heart of the regulatory debate.
AI Asia Business Companies Technology World