Read Dario Amodei's post "Our position on open-weights models" (Anthropic, Jul 27 2026). Key points:
- Anthropic has never advocated banning open-weights models. Open models w/o dangerous capabilities are a public good.
- Real concerns are (1) authoritarian govts (esp. CCP) building more powerful models for military/surveillance dominance, and (2) misuse for cyber/bio attacks + alignment risks — neither is solved by banning US businesses from using open models.
- What Anthropic actually supports: (1) keep advanced chips/chipmaking equipment out of China + crack down on smuggling, (2) crack down on industrial-scale distillation operations, (3) mandatory safety testing for all sufficiently capable models, open or closed.
- Pushes back on the industry open letter's claim that open weights inherently help defenders more than attackers — argues this should be settled by empirical testing, not assumed (esp. relevant for bio risk where attack may be easier than defense).
Good reminder that the "open vs closed" framing oversimplifies the actual policy levers that matter.