Anthropic’s Dario Amodei responds: doesn’t oppose open-weight models, but fears Chinese AI
TechCrunch Julie Bort ● Covered by 75 sources
Dario Amodei says Anthropic never pushed to ban open-weight AI models, despite industry chatter. His real worry: authoritarian states, especially China, building unchecked, more powerful AI.
Based on reporting by TechCrunch, Julie Bort — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Dario Amodei spent Monday afternoon clearing his name. The Anthropic CEO wrote a blog post flatly denying that his company has ever advocated banning open-weight models, Chinese or otherwise, after days of industry murmuring suggested otherwise. "Anyone who has read my past writing should know that I don't regard such bans as a useful measure," he wrote, italics his.
The timing wasn't random. Nvidia's Jensen Huang used his very first post on X last Friday to share an open letter signed by Nvidia, Hugging Face, Meta, Microsoft, and Mistral, among others, warning policymakers against "premature restrictions" on open-weight AI. The letter never mentioned China by name, but everyone in the room knew the subtext: fears that Chinese labs are catching up fast, partly through distillation, where a model is peppered with prompts until it essentially copies how another model thinks.
Amodei's actual position is more nuanced than a simple pro- or anti-open-source stance. He called open-weight models without dangerous capabilities a genuine public good, cheap to run and useful to developers and researchers. His concern isn't the technology's openness so much as what happens when it lands in the hands of authoritarian governments. He named the Chinese Communist Party specifically as the most capable of the authoritarian actors he worries about, chasing what he called permanent military superiority, alongside the domestic use of AI for repression and the risk of biological attacks.
What makes open-weight models riskier in his eyes is that once they're released, there's no putting them back in the box. He cited a UK AI Security Institute report to back that point, and argued it's much harder to slap guardrails on a model or track its use once the weights are public. That's a direct rebuttal to the open-source camp's usual argument, that widely available models actually help defenders more than attackers.
As for solutions, Amodei's list reads like familiar policy: keep restricting China's access to advanced chips, and push for a real crackdown on distillation, with the U.S. already threatening sanctions over alleged IP theft. But the more interesting bit is his openness to a global model safety testing body, one China would also have to join. He seems to think that's not far-fetched, floating the idea that even Beijing might see value in cooperating on preventing AI-enabled biological weapons, since the risk cuts both ways.
My take — AI-written commentary, not fact-checked reporting
Amodei is trying to have it both ways here, and mostly pulling it off. He gets to sound like an open-model ally while still making the case for tighter chip controls and a distillation crackdown aimed squarely at China. Fine, but the idea that a global safety-testing regime could actually rope in the CCP feels like wishful thinking dressed up as pragmatism, and it conveniently sidesteps the harder question of who gets to decide which models count as dangerous in the first place.
Read more about this at: TechCrunch