Anthropic denies wanting an open-weights ban
Dario Amodei says Anthropic never asked for a blanket ban on open models, and instead wants chip controls, distillation limits and mandatory safety tests.
Anthropic published a policy statement on July 27 stating plainly that it has “never advocated for a ban on open-weights models” as a category, an explicit response to a narrative that has followed the company through this year’s open-weights debate. CEO Dario Amodei writes that models without dangerous capabilities cost nothing beyond compute to run and provide real value to businesses, developers and researchers who couldn’t otherwise afford frontier access.
What Anthropic does support instead: chip export controls that limit China’s access to advanced hardware, on the reasoning that domestic production capacity there still lags; a crackdown on “industrial-scale distillation operations” that let a lagging lab close the capability gap by training on a frontier model’s outputs rather than its own research; and mandatory pre-release safety testing, for open and closed models alike, covering cyber, biological and alignment risks. Amodei frames the underlying concern as two “nightmare scenarios,” an authoritarian government building superior military AI, or a powerful model being misused for cyberattacks or bioweapons design.
What it means for operators
This lands the same week Moonshot released Kimi K3’s weights openly and Anthropic itself pursued a distillation complaint against Alibaba, so the position statement reads less like abstract philosophy and more like Anthropic drawing a line around a live fight. If your model-selection strategy leans on open weights for cost or control, the actual regulatory risk to watch isn’t a ban, it’s how far chip-export and distillation rules eventually reach into which specific open models can legally train on which specific outputs, since that is the exact policy Anthropic is now on record pushing for.