"No ban on open models, Anthropic asserts"

Anthropic denies any desire to ban open models. Dario Amodei proposes targeting physical chips and industrial distillation.

After several days of criticism accusing it of seeking to ban open-source models to protect its business, Anthropic has outlined its position on the matter, penned by its CEO Dario Amodei. His central statement serves as a denial: the company has reportedly never advocated for such a ban, and in its view, a model devoid of dangerous capabilities constitutes a public good.

The text responds to the letter Open Weights and American AI Leadership, which Anthropic did not co-sign. Amodei says he shares the view that open models expand access, competition, and customer control, but rejects the argument that openness makes AI safer or gives defenders an advantage over attackers. His concern, he specifies, is not with openness itself, but with frontier models reaching a level that can be exploited for military, cyber, or biological purposes once released with no possibility of recall.

Rather than a ban, which it deems inadequate to address its national security concerns, the company proposes three measures: tightening controls on advanced chips and manufacturing equipment destined for authoritarian governments, action against industrial-scale distillation, and mandatory safety testing for any sufficiently capable model, whether open or closed.

Regarding distillation, Anthropic points back to an accusation already made before a U.S. Senate committee: Alibaba's Qwen laboratory reportedly conducted the largest known operation of this kind against Claude, using tens of thousands of fake accounts. Amodei acknowledges that such a practice is difficult to curb, as accounts are often only detectable once the distillation is well underway, and calls for a regulatory response rather than leaving it to the initiative of a single company.