On July 27, 2026, Anthropic published “Our position on open-weights models,” a statement signed by CEO Dario Amodei1. In it, Amodei states plainly that “Anthropic has never advocated for a ban on open-weights models,” and stresses repeatedly that the company is not seeking any regulation that would prohibit open-weight models as a category1.
The statement lands in the middle of an escalating debate over how open-weight AI should be regulated. On July 24, NVIDIA joined Hugging Face, Meta, Microsoft, Mistral, and many other AI companies in an open letter urging policymakers not to impose broad “premature restrictions” on open-weight models — a letter Anthropic notably did not sign2.
The Substance: Chips, Distillation, and Safety Testing Instead of a Ban
Amodei frames open-weight models without dangerous capabilities as a public good2. His central worry, he explains, is authoritarian governments building models more capable than America’s for military superiority or population control12. From there, the post narrows Anthropic’s supported policy interventions to three1.
First, the United States should not sell powerful chips or chipmaking equipment to China. Amodei argues that “China has limited domestic production capacity, and therefore, due to the scaling laws, cannot build more powerful models than the US without US chips,” making export controls the lever that addresses the root of the threat1.
Second, governments should crack down on industrial-scale distillation operations. Distillation — training a new model on the outputs of an existing one — is “a much more compute-efficient process than training models from scratch,” and Amodei warns it could “bring the Chinese frontier to within a few months of the US frontier”1. For a primer on how the technique works, see our explainer on model distillation and quantization.
Third, all sufficiently capable models — open and closed alike — should go through mandatory safety testing for cyber, biological, and alignment risks before release1.
Why Anthropic Didn’t Sign the Industry Letter
Advocates of open weights often argue that broad availability accelerates safeguards research and strengthens defenders. Amodei pushes back on that premise: he does not agree that open-weights models necessarily make it easier to develop safeguards, and questions whether broad access to capabilities necessarily helps defenders more than attackers1. The line he draws is that Anthropic is not hostile to open release itself, but declines to endorse the standard defense of it.
According to TechCrunch, there had been murmuring in the industry that Anthropic somehow supported U.S. government efforts to ban open-weight Chinese models — or possibly open-weight models in general — and the new post responds to that2. It reads as a direct answer: not a ban, but targeted interventions on chips, distillation, and safety testing.
The Backdrop: One Turbulent Week After the Moonshot Accusations
The statement caps a frenetic stretch of news. On July 22, White House OSTP director Kratsios publicly accused China’s Moonshot AI of secretly distilling Anthropic’s model to build Kimi K3, and Treasury Secretary Bessent floated possible sanctions (see our report on the distillation accusations and sanctions warning). Five days later, on July 27, Moonshot went ahead and released the full weights of Kimi K3 — at 2.8 trillion parameters, the largest open-weight model ever published.
Within a single week, the U.S. government’s distillation accusations, the industry’s open letter, and a record-setting open-weight release all collided — and Anthropic’s official position arrived as the closing statement. Notably, the post itself names no specific Chinese model, framing its concerns about Chinese open-weight models in general terms1.
The Axis of the Debate Is Shifting
For readers weighing AI adoption decisions, the statement is a useful signal of where open-weight regulation is actually heading. Even Anthropic, having now published this position, is not asking for restrictions on the release format itself. The policy conversation is consolidating around supply-chain and verification controls: chip exports, distillation, and safety testing. Nothing in the current debate suggests that using open-weight models will be restricted in the near term.
That said, if distillation crackdowns or mandatory safety testing become law, scrutiny of where a model comes from will only intensify. Enterprises considering Chinese-origin open-weight models like Kimi K3 will need to weigh geopolitical risk and regulatory exposure alongside performance and licensing. Which of the three proposals becomes actual policy is worth tracking — together with the EU AI Act’s transparency obligations taking effect August 2, it will shape the near-term direction of AI regulation.
Sources
- Our position on open-weights models - Anthropic official post by CEO Dario Amodei
- Anthropic’s Dario Amodei responds: doesn’t oppose open-weight models, but fears Chinese AI - TechCrunch