Dario Amodei, founder and CEO of Anthropic he replied on Monday afternoon in the industry to say that his company somehow supports the US government’s efforts to ban Chinese bantamweight models, or bantamweight models in general.
“Anyone who has read my past writings will know that I do not consider such bans to be a useful measure, but to be clear, let me state clearly: Anthropic has never advocated banning overweight models“, he emphasizes himself by writing.
His answer came from Nvidia founder and CEO Jensen Huang X (his first post on the platform) to share An open letter on Friday Nvidia and a long list of other AI companies, including Hugging Face, Meta, Microsoft, Mistral and Nvidia, urged policymakers not to impose broad “premature limits” on open-weight AI models.
Although China was not specifically mentioned in that letter, the debate in the industry centered on allegations that Chinese AI labs often boosted their capabilities by stealing intellectual property from their American counterparts. One method is distillation, where the AI bombards another model to learn how it works.
Amodei said in a blog post published Monday that the lightweight models fall into a different bucket than the threat posed by China. “Open weight models with no risk capabilities are a public good: they cost nothing other than the computation needed to run them, and provide value to businesses, developers, and researchers.”
In his post, Amodei said there have long been fears about artificial intelligence, but enterprises using open weight models, even those from China, are not among them.
His fear is that “authoritarian governments” will build stronger models than those in the US to use AI to gain “permanent military advantage” and/or repress their own people. He said the Chinese Communist Party (CCP) was not the only authoritarian government he was worried about, but it was “the most capable”.
Amodei said he fears AI will enable not only cyber security but also “biological attacks”. And in his opinion, lightweight models are more dangerous in these scenarios because they are difficult to apply guardrails or control their use. He also cited a report by the UK’s AI Security Institute that once open weights are issued, they cannot be withdrawn.
This is contrary to what open source advocates claim: access to robust open models that are not controlled by a single entity helps advocates protect themselves.
Amodei listed actions he believed would prevent China, including limiting access to powerful chips (a long-standing US policy) and calling for a formal crackdown on distillation. The US threatened Sanctions against China If it detects IP theft related to US models.
Interestingly, Amodei also expressed his support increasing efforts, some led by the United States, Establishing a model safety testing organization, especially if the whole world agrees to follow it, including China.
“I think this idea is actually close to consensus: I’m very pleased with both the Trump administration’s move in this direction in recent months and the industry proposals to apply this type of testing to the most capable models, regardless of country of origin or whether they’re open or closed.
“Note that in order to be effective,” he added, “the test must be global, which means it must also be in the CCP.” I think it might actually be possible … limited cooperation on preventing AI bioweapons might be possible because that’s in China’s interest as well.”
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
