Does Anthropic support banning open-weight AI models?

No. In a July 27 statement, CEO Dario Amodei said Anthropic “has never advocated for a ban on open-weights models” and instead wants chip controls, curbs on distillation, and mandatory safety testing regardless of a model’s license.

What prompted Amodei's clarification?

An industry letter, “Open Weights and American AI Leadership” , that Nvidia’s Jensen Huang published on July 24 and that grew to roughly 50 signatories — without Anthropic — fueled speculation the company wanted Chinese open-weight models banned outright.

What is AI model distillation, and why does Amodei want it restricted?

Distillation means training a new model on another model’s outputs to cheaply copy its abilities. Amodei argues unchecked, industrial-scale distillation could let Chinese labs close the gap with US frontier AI “within a few months.”

Does Amodei oppose open-weight models in general?

No — he called smaller open-weight models without dangerous capabilities “a public good” for research and competition, and says the line should be drawn only at frontier-level systems capable of serious harm.

Anthropic's Amodei Denies Seeking Open-Weight AI Ban

Anthropic chief executive Dario Amodei used a July 27 company blog post to reject the idea that Anthropic wants open-weight AI models banned, pushing back on speculation that had built up after a rival industry letter circulated the week before.

“Anthropic has never advocated for a ban on open-weights models,” Amodei wrote, according to the statement.

The clarification follows an open letter, “Open Weights and American AI Leadership,” that Nvidia CEO Jensen Huang first posted on July 24 and that grew from roughly 25 to about 50 signatories within a day, urging Washington not to restrict open-source AI development. Anthropic did not sign, feeding assumptions that the company was quietly pushing for a ban on Chinese open-weight systems.

Amodei’s post draws a line between the kind of open model he considers “a public good” — one without dangerous capabilities — and the much narrower set of frontier systems he worries about. He argues safety obligations should depend on what a model can actually do, not whether its weights are open or closed, and that open access does not automatically make software safer or hand cyber defenders an advantage over attackers, as some open-model advocates claim.

Chips, distillation and testing

Instead of a ban, Amodei laid out three specific requests for policymakers. First, keep advanced chips and chipmaking tools out of China and crack down on smuggling, since he says China’s own manufacturing capacity still can’t match US supply. Second, restrict industrial-scale distillation — training a new model by mining another model’s outputs — which he argues could otherwise let Chinese labs close the gap with the US frontier within months. Third, require mandatory safety testing, covering cyber, biological and alignment risks, for any sufficiently capable model regardless of whether it is open or closed.

Underlying the statement is Amodei’s concern that an authoritarian government could use a distilled or domestically built frontier model to gain lasting military advantage. He also floated the idea of an international AI safety-testing body that includes China, saying he sees room for cooperation specifically on preventing AI-assisted biological weapons work.

The exchange adds to a broader Washington regulatory fight over how to handle Chinese open-weight releases such as Moonshot AI’s Kimi K3, with chipmakers and cloud providers pushing for openness while safety-focused labs like Anthropic argue that capability, not licensing, should determine the rules.

Chips, distillation and testing

Frequently asked questions