Anthropic’s Amodei Rejects Ban on Open-Weight Models, Draws Line at Chinese Distillation
Anthropic CEO Dario Amodei published a direct rebuttal Monday to accusations that his company wants open-weight AI models banned, insisting Anthropic has never pushed for such a ban even as Moonshot AI's free, frontier-competitive Kimi K3 model reshapes the debate.
Dario Amodei’s post arrived after days of industry chatter accusing Anthropic of quietly favoring restrictions on open-weight models as a way to protect its own closed, subscription-based business.
He shut that down directly, writing in his own emphasis that Anthropic has never advocated for a ban on open-weight models, and that AI models lacking dangerous capabilities cost nothing beyond compute while providing real value to businesses, developers, and researchers.
A Letter Anthropic Refused to Sign
TechCrunch reported that Amodei’s response followed Nvidia CEO Jensen Huang’s first-ever X post, an open letter signed by Nvidia, Hugging Face, Meta, Microsoft, Mistral, and other companies urging policymakers not to impose broad restrictions on open-weight models.
Axios noted that by Monday, Anthropic stood out as the most prominent holdout from that campaign, even as Google, Palantir, and ChatGPT creator OpenAI all eventually lined up behind it.
Amodei didn’t dispute the letter’s core claims about access and competition, agreeing that open models expand participation in the AI economy.
However, he pushed back on its assertion that broad access to model weights necessarily helps defenders more than attackers, calling that assumption unproven rather than obviously true.
Moonshot’s Kimi K3 Forced the Issue
The immediate trigger for the industry-wide letter was the release of Kimi K3, a 2.8-trillion-parameter open-weight model from Chinese lab Moonshot AI.
It landed on Hugging Face and immediately posted benchmark scores rivaling Anthropic’s own Fable 5 and OpenAI’s GPT-5.6 Sol, all available for free download.
Anthropic’s own post pointed to this exact dynamic as the source of confusion.
Amodei clarified that his concern was never about open weights themselves, but about authoritarian governments using AI to control their populations or gain military superiority, a stance that caused legal tensions between Anthropic and the Pentagon.
He added that this was a concern he had held consistently for years, rather than one invented to justify restricting a Chinese rival.
Chips, Distillation, and Testing Instead of a Ban
Rather than supporting any blanket restriction, Amodei called for three specific interventions. These included tighter controls on advanced chips and chipmaking equipment flowing to authoritarian governments.
Secondly, he called for a formal crackdown on industrial-scale distillation, where one model is trained by extensively querying a more advanced rival. Finally, he urged mandatory safety testing for all sufficiently capable models, regardless of whether they’re open or closed.
That third point connects to a broader complaint Anthropic has raised before; the company told the U.S. Senate Committee last month that Chinese firm Alibaba had carried out what it called the largest known distillation attack against a Western AI lab.
Amodei explicitly rejected the idea that a ban would solve the problem he was worried about. He argued that legitimate U.S. businesses were never the threat.
Instead, the greater risk comes from models trained in secret and handed directly to state military or surveillance agencies, rather than models businesses can openly evaluate and adopt.
