Anthropic CEO: Never Advocated Banning Open-Weight AI
Claire Weston
Anthropic CEO Dario Amodei published a blog post declaring the company has never advocated banning open-weight models — yet Anthropic still refuses to sign a letter backed by 20-plus tech giants, signaling he is trying to thread a very narrow needle between embracing openness and insisting the most powerful models need guardrails.
What pressure is he responding to?
Nvidia, Microsoft, Meta, Google, OpenAI, and over 20 other tech companies signed an open letter urging the US government not to restrict open-weight AI.
Anthropic has not signed. It is the most conspicuous holdout among major AI labs.
White House AI adviser David Sacks publicly accused Anthropic of using safety concerns to protect its own business — This means → the pressure comes not just from peers but from the White House itself.
What is Amodei actually saying?
He acknowledged that banning Chinese open models "would insulate Anthropic from competition," but stressed "that has never been my goal."
He classified open models without dangerous capabilities as "a public good" — In plain terms = he is fine with ordinary open-source models and even sees them as beneficial.
But he pushed back on the letter's core claim: open-weight models do not necessarily make AI safer, nor do they inherently give defenders an edge over attackers.
If not an outright ban, what does he want?
Amodei proposed three targeted alternatives: tighter controls on advanced chips flowing to authoritarian governments; a crackdown on "industrial-scale distillation" — training a model on the outputs of a stronger one, essentially learning by copying answers; and mandatory safety testing for sufficiently capable models, whether open or closed.
He argued a blanket ban "would not address my most serious national-security concerns" because bad actors are unlikely to be legitimate US companies.
This reflects his real position: the issue is not "open vs. closed" but at what capability threshold regulation must kick in.
Why is Kimi K3 the catalyst?
Moonshot AI's Kimi K3, a Chinese open-weight model, approached top US model performance at a fraction of the cost — sending a shockwave through Silicon Valley.
The Trump administration had already been debating action against Chinese open-source models, including sanctions on labs accused of using distillation to steal US AI research.
This means → Amodei's timing was no accident — the Chinese-model shock forced every player to show its hand.
Can this tightrope act hold?
Google and OpenAI initially held off but joined the letter last weekend; Anthropic remains the only major closed-source lab that has not signed.
The line Amodei is trying to draw: oppose a blanket ban, but insist on oversight for the most powerful AI systems.
In plain terms = he wants to avoid being cast as anti-open-source while refusing to abandon his core belief that strong models must be regulated — whether the industry buys it remains to be seen.
Content is for reference only, not financial advice.