Anthropic says it does not want open-weights models banned

Anthropic set out its official view on open-weights models, and the short version is that it does not want them banned. Dario Amodei writes that the company has never argued for a ban, and that open models without dangerous capabilities are a public good. The worry he puts first is not open source itself but a strategic one: that an authoritarian government, China in particular, ends up with models stronger than anything the United States has, and uses that lead for military advantage or domestic control. Misuse for cyber and biological attacks comes second, and here open weights are harder to manage, because you cannot add guardrails after release and you cannot take the weights back.

Instead of bans, Anthropic argues for three narrower controls. Keep advanced chips out of China through export rules and by shutting down smuggling. Crack down on industrial-scale distillation, which Amodei says can let Chinese labs train models within a few months of the US frontier despite limited chip access. And require pre-release safety testing for cyber, biological, and alignment risk on any sufficiently capable model, open or closed, applied to Chinese models too. The post also breaks with an open-weights industry letter, rejecting its claim that open models automatically help defenders more than attackers. You can read the full position here.

Why it matters

If you build on open-weight models, the regulatory fight is not really about whether you can download weights. It is about chips, distillation, and mandatory testing, which are the levers that could change which models exist and what you are allowed to ship.