Anthropic chief executive Dario Amodei denied that his company ever sought a ban on open-weight AI models, and named three policy steps he backs instead.
Key Points:
- Amodei said Anthropic has never advocated a ban on open-weights models as a category.
- He wants chip export controls, a distillation crackdown and mandatory safety testing for capable models.
- Anthropic was absent from an open letter signed by more than 20 technology companies.
Amodei Denies Open-Weight Ban Push
Amodei posted the statement on Jul. 27, ending several days of silence that rivals across the industry had treated as an answer in itself.
He called open models without dangerous capabilities a public good, saying they cost nothing beyond the compute needed to run them and deliver value to businesses, developers and researchers.
Protectionist bans would not touch his most serious national security concerns, he added.
His post answered an open letter that Nvidia chief Jensen Huang circulated on Jul. 24. More than 20 companies, among them Microsoft, Meta and Palantir, signed the letter, which Huang shared in his first message on X.
Also Read: Crypto Wrench Attacks Reach 52 Cases In Six Months, Most Of Them In France
Sacks And Gurley Question Anthropic Motives
David Sacks, a venture capitalist and former White House AI czar, argued that Anthropic was invoking safety language to defend its own business model. Bill Gurley, a partner at Benchmark, suggested the company was guarding its place in the race rather than making a principled case. Anthropic was the last large American lab left outside the letter once OpenAI signed.
A ban would shield American labs from competition, Amodei acknowledged, but that was never his goal. He wants tighter chip export controls, a crackdown on industrial-scale distillation and mandatory safety testing for every sufficiently capable model, open or closed.
He splits his fears in two.
The first is authoritarian governments, with the Chinese Communist Party the most capable among them, building models stronger than American ones and using them for lasting military superiority or repression at home. The second is misuse of capable models for cyberattacks or biological attacks, a risk he says open weights sharpen because guardrails come off easily and released weights cannot be recalled.
Kimi K3 Claims Fuel Distillation Fight
Chinese lab Moonshot AI released Kimi K3 in mid-July, and the open model matched leading American systems on several benchmarks at a fraction of their price.
A senior White House official then accused Moonshot of training it on outputs from Anthropic's Fable model, a practice known as distillation, and Treasury Secretary Scott Bessent floated sanctions in response.
Amodei has held this line for years. He set out both scenarios in an essay published six months ago, and he has warned before that developers lose control of a model once its weights become public. Anthropic has leaned on gated distribution instead, including Project Glasswing, which gives vetted companies early access to its cyber models.
Read Next: XRP Near-Term Forecasts Top Out At $1.25 Before The Fed Decision