The story

Anthropic CEO Dario Amodei said in a blog post on Monday that his company has never advocated for a ban on open-weight models, attempting to address criticism after Anthropic was conspicuously absent from a letter signed by over 130 companies including Nvidia, Microsoft, Meta, OpenAI, and Google. The letter, released late last week, urged policymakers to avoid premature restrictions on open-weight models, which users can download, modify, and run on their own infrastructure. Chinese startups currently dominate the market, and some government officials have started to weigh whether those models should be banned or restricted in the U.S.

Amodei wrote that Anthropic should instead focus on keeping powerful chips out of authoritarian hands, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models, open and closed. Distillation is an AI training method where a smaller model is built using outputs from a bigger model already on the market. Anthropic sent a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs last month alleging that China's Alibaba had carried out the largest known distillation attack against it to date. While Amodei agreed that open-weight models give customers greater control and expand access to the AI economy, he disputed the open letter's assertions that such models necessarily make it easier to develop safeguards or help defenders more than attackers.

Chinese companies including Moonshot, DeepSeek, and Z.ai are releasing models that are close to the performance of the best American systems but at a fraction of the price. Vercel CEO Guillermo Rauch says open-weight models accounted for about 55% of tokens flowing through his company's AI Gateway in July, up from just 4% in January. The shift accelerated after the releases of DeepSeek V4, Z.ai's GLM-5.2, and Moonshot's Kimi K3. William Blair analysts argued recently that cheaper, more open AI models are unambiguously good for software companies because AI is simply another input cost.