Anthropic open-weight model ban stance clarified by Dario Amodei
Dario Amodei said Anthropic does not back a category ban, as major tech firms press policymakers to protect open-weight AI models.
By Sarah Jenkins · Chief Macro Economics Correspondent
· 3 min read
Anthropic CEO Dario Amodei said the company is not seeking an Anthropic open-weight model ban, moving to clarify its policy position as a widening group of technology companies warns Washington against broad limits on downloadable AI systems. The dispute affects companies building and buying AI infrastructure, including Nvidia, Microsoft, Meta and Palantir, because open-weight models can be run outside a developer’s own cloud services and can shape demand for chips, software and enterprise tools.
In a blog post published Monday, Amodei wrote that Anthropic has “never advocated for a ban on open-weights models” as a category. He said Anthropic’s preferred focus is restricting access to advanced chips by authoritarian governments, preventing large-scale model distillation and requiring safety testing for sufficiently capable systems, whether open or closed.
The statement followed a letter released late last week by a coalition that included Nvidia, Microsoft, Meta and Palantir. The companies urged policymakers to avoid “premature restrictions” on open-weight AI models, according to the letter. OpenAI later added its name to the letter, while Anthropic did not.
What is an open-weight AI model?
An open-weight model is an AI system whose core parameters can be downloaded, modified and run on a user’s own infrastructure. That structure differs from closed models, such as Anthropic’s Claude family, where customers typically access the system through services controlled by the developer.
Supporters of open-weight systems say the model gives customers more control, expands access to AI tools and can increase competition in certain use cases. Amodei said in his blog post that he agrees with those points in the industry letter.
The debate has taken on geopolitical weight because Chinese startups currently dominate the open-weight model market, and some U.S. government officials have been considering whether those systems should face bans or restrictions in the United States, CNBC reported. Anthropic sells proprietary models to businesses, while OpenAI also primarily develops closed models.
Why Anthropic is focused on chips and distillation
Amodei said bans aimed at protecting domestic companies would not address what he described as Anthropic’s most serious national security concerns. His blog post instead emphasized controls on powerful chips, which are used to train and run advanced AI systems, and safeguards against distillation.
Distillation is a training technique in which a smaller model is built with help from the outputs of a larger model already in use. Anthropic told the U.S. Senate Committee on Banking, Housing, and Urban Affairs last month that Alibaba, the Chinese company behind the Qwen model family, had conducted what Anthropic called the “largest known distillation attack” against it to date.
The industry letter said concerns about unlawful distillation should be handled through targeted legal and commercial frameworks. Amodei said Monday that he agreed with that approach.
He also set out areas of disagreement with the companies that signed the letter. Amodei said he does not accept the view that open-weight models favor cyber defenders over attackers, and he disputed the claim that open-weight access makes it easier for users to develop safeguards.
Even with those reservations, Amodei’s blog post drew a clear boundary around Anthropic’s position. The company is arguing for safety testing and targeted controls, while saying it does not support a blanket prohibition on open-weight AI models.
This story draws on original reporting from CNBC.
