Anthropic Doesn't Want Open Weight Models Banned. Just All That Makes Them Good
9 hours ago
- The US government is considering banning or limiting Chinese open-weight AI models, citing national security and distillation concerns.
- Nvidia led an open letter supporting open-weight models, which many AI companies signed, but Anthropic, OpenAI, and Google initially did not; OpenAI and Google later signed, leaving Anthropic as the holdout.
- Anthropic's Dario Amodei says the company opposes a blanket ban on open-weight models but advocates for policies that would restrict distillation, tighten chip export controls, and mandate pre-release safety testing—measures critics argue would cripple competitive open models.
- Distillation, the method of training a smaller model using outputs from a stronger one, is labeled as theft by Treasury Secretary Bessent, but the article argues it is a form of learning, not stealing, and that similar practices were used by frontier labs themselves.
- Open-weight models foster competition, give customers control, and allow broader security testing, whereas relying solely on closed models creates single points of failure and concentrates power.
- Anthropic's stance is seen as protectionist, using safety arguments to justify hobbling cheaper competitors while its own closed models benefit from reduced competition.