Amodei: Anthropic never backed an open-weights model ban
Anthropic CEO Dario Amodei says the company has never backed a ban on open-weights models, calling for chip controls on China and mandatory safety testing.
In Brief
- Anthropic CEO Dario Amodei wrote that the company has “never advocated for a ban on open-weights models” as a category.
- Anthropic backs chip export controls and a crackdown on “rampant smuggling” of advanced chips into China instead of a ban.
- Amodei wants mandatory safety testing for all sufficiently capable models, open and closed, with smaller models exempted.
Anthropic CEO Dario Amodei says the company has “never advocated for a ban on open-weights models,” pushing back on weeks of speculation that the lab wants to choke off open AI. The clarification, covered by CNBC and TechCrunch, landed after a coalition letter urged policymakers against restrictions.
The debate ignited after Nvidia’s Jensen Huang and dozens of companies, including Hugging Face, Meta, and Microsoft, signed an open letter opposing “premature restrictions” on open-weight models. OpenAI later signed; Anthropic did not, and its silence drew fire.
Amodei framed his position as narrow: not against open weights, but against open weights built by stealing from closed models, trained on smuggled chips, and shipped without safety testing. Moonshot Kimi K3 shows how fast open models are moving.
Open-weights models are a public good
Amodei wrote that open-weights models without dangerous capabilities are “a public good,” echoing language from The Next Web’s coverage of his post. He agreed with parts of the industry letter: open weights expand access, strengthen competition, and give customers control.
He broke with open-source advocates on one claim, arguing open weights do not necessarily help defenders more than attackers, and do not make safeguards easier. Once released, he noted citing a UK AI Security Institute report, open weights cannot be withdrawn.
His sharpest worry is biological: a sufficiently capable model could weaponize a pandemic-level virus using widely available materials, while defense takes years. That risk, he said, is why testing matters more than bans for the most capable systems.
Chip controls target China
Amodei’s preferred tools are old and new: enforce chip export controls and crack down on the “rampant smuggling” moving advanced chips into China, because without U.S. silicon, scaling laws cap what China can train. He also wants a formal crackdown on distillation.
Anthropic has accused Alibaba’s Qwen lab of the largest distillation campaign ever against Claude, using 25,000 fake accounts across 29 million exchanges. Amodei called policy intervention necessary because accounts are usually spotted only after the damage is done.
He also endorsed a global model-safety testing body, even one that includes China. “Protectionist bans would not address my most serious national security concerns,” Amodei wrote, pointing to limited cooperation on AI biological weapons as a place both sides share interest. Hugging Face transparency advocates make a similar case for openness with guardrails.
FAQ
Did Anthropic call for a ban on open-weight models?
No. Dario Amodei wrote that Anthropic has “a ban on open-weights models” concerns and argued protectionist bans would not address his national security concerns.
What does Anthropic want instead of a ban?
Anthropic backs chip export controls, a crackdown on distillation and chip smuggling, and mandatory safety testing for all sufficiently capable models, open and closed.
Why is Anthropic focused on China?
Amodei says scaling laws mean China cannot train more powerful models without U.S. chips, and he alleges Alibaba’s Qwen ran the largest distillation attack on Claude to date.
