Anthropic Rejects Calls for a Ban on Open-Weight AI Models

Anthropic CEO Dario Amodei said the company has never supported banning open-weight AI models, as Washington debates potential restrictions on Chinese open-weight models

In a blog post, Amodei said reports suggesting that some US officials were considering banning Chinese open-weight models had led to claims that Anthropic supported such measures to protect its commercial interests. 

“Anthropic has never advocated for a ban on open-weights models,” he wrote. “Open-weights models that don't have dangerous capabilities are a public good.”

The statement comes as more than 50 technology companies and organisations, including NVIDIA, Microsoft, Meta, IBM, Dell Technologies and Hugging Face, signed an open letter opposing restrictions on open-weight AI models and arguing that they are important for innovation, competition and national security.

Instead of restricting open-weight models, Amodei outlined three policy measures that he believes would better address national security risks.

To begin with, he called for tighter controls on exports of advanced AI chips and chipmaking equipment to China, along with stronger enforcement against smuggling. According to Amodei, China’s limited domestic semiconductor manufacturing capacity means access to advanced US chips remains a key factor in training frontier AI models.

He also urged governments to target what he described as “industrial-scale distillation operations”. Distillation is a training technique that allows smaller models to learn from larger ones while requiring far less compute than training from scratch.

“Distillation is a much more compute-efficient process than training models from scratch,” Amodei wrote. “It allows China to build much better models than its number of chips would ordinarily enable.”

He argued that policies should focus on deterring state-backed distillation efforts rather than banning open-weight models. “A blanket ban on open-weight models is neither the correct remedy nor something we have called for,” he said.

Lastly, Amodei proposed mandatory safety testing for all sufficiently capable AI models, regardless of whether they are open or closed. He said testing should evaluate cyber, biological and alignment risks before deployment.

“The best way to address threat #2 is to just directly test models for cyber, biological, and alignment risks before release,” he wrote, adding that the approach should apply equally to open and closed models while exempting less capable models from startups and academia.

Amodei also addressed arguments made in the industry-backed open letter. While agreeing that open-weight models broaden access to AI and strengthen competition in some areas, he disputed claims that they inherently improve security.

“I don't agree with the letter’s assertions that open-weight models necessarily make it easier to develop safeguards or that broad access to capabilities necessarily helps defenders more than attackers,” he wrote. “It seems at least as likely to me that the opposite will be true.”

He warned that advances in biology could create an attacker advantage if capable AI models make it easier to develop dangerous pathogens while defensive measures continue to take years to deploy.

Summarising Anthropic’s position, Amodei said the company’s focus remains on “keeping powerful chips out of authoritarian hands, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models, open and closed,” rather than banning open-weight AI models.