Dario Amodei (Anthropic) denies supporting a ban on open-weight models, warns of AI misuse by authoritarian governments

Anthropic chief Dario Amodei publicly responded on Monday to speculation that his company supports US government efforts to ban open-weight models, either in general or specifically Chinese models. In a blog post, he wrote that Anthropic has never advocated a ban on open weights and does not consider such bans a useful measure. He was responding to an open letter shared on X on Friday by Nvidia founder Jensen Huang and signed by a number of companies, including Hugging Face, Meta, Microsoft and Mistral — the letter urged policymakers not to introduce “premature restrictions” on open-weight models.

According to Amodei, open-weight models without dangerous capabilities are a public good because they cost nothing beyond the computing power needed to run them and provide value to companies, developers and researchers. In his view, the real threat is not open models themselves, but a scenario in which “authoritarian governments” — among which he singles out the Chinese Communist Party as the most capable — develop more powerful models than the US and use them to achieve lasting military superiority or suppress their own populations. He also mentioned concerns about AI-enabled biological attacks and referred to a report by the UK AI Security Institute stating that open weights, once released, cannot be withdrawn, which he says makes it harder to apply safety measures and monitor their use. This view conflicts with the argument made by open source advocates that access to powerful open models that are not controlled by a single entity actually helps defend against attacks.

As specific measures targeting China, Amodei cited restrictions on access to powerful chips (a longstanding US policy) and formal penalties for distillation, a technique in which one model acquires knowledge by repeatedly querying another model. He also said he supports growing efforts, led in part by the US, to establish a global organization for model safety testing — on the condition that the entire world, including China, would participate. He stated that, in his view, the administration of President Trump has already moved in this direction and that limited cooperation with China on preventing AI-based biological weapons is possible because it is also in the interest of China.