Analysis
Anthropic has become the sole notable holdout among frontier AI labs on a Nvidia-backed open letter urging the US government not to restrict open-weight AI models, a category where Chinese developers currently lead. The company's absence from a list that includes most of its major competitors has drawn public scrutiny, with Axios framing Anthropic's position as increasingly isolated -- "Anthropic's lonely island" -- within an industry that otherwise broadly favors keeping open-weight development unrestricted.
CEO Dario Amodei addressed the criticism directly in a blog post, clarifying that "Anthropic has never advocated for a ban on open-weights models" and describing models without dangerous capabilities as "a public good." That's a meaningfully softer position than critics initially assumed from Anthropic's refusal to sign the letter. Amodei's actual argument is more specific: banning US companies from using Chinese open-weight models would do little to stop determined bad actors or hostile governments from accessing them regardless, making a blanket ban both ineffective and potentially counterproductive.
Instead of a ban, Amodei's proposed approach targets the inputs China needs to keep closing the AI gap: keeping advanced chips and chipmaking equipment out of the country, pursuing smuggling networks that circumvent existing export controls (a live issue given this week's separate Nvidia-Moonshot-Thailand allegations), and cracking down specifically on industrial-scale distillation -- the technique that lets Chinese labs train competitive models by learning from US frontier model outputs without matching the underlying compute investment.
“The criticism Anthropic is facing isn't really about the substance of its position -- it's about incentives.”
The criticism Anthropic is facing isn't really about the substance of its position -- it's about incentives. White House AI adviser David Sacks has publicly suggested Anthropic is using safety rhetoric to protect its own commercial model, since Anthropic runs a closed, proprietary Claude business that benefits directly from any policy environment that constrains open-weight alternatives, whether American or Chinese. That's a difficult accusation for Anthropic to fully rebut, since its business interests and its stated safety concerns do genuinely point in the same policy direction, even if the underlying concerns are sincere.
This compounds an already complicated week for Anthropic's positioning: the company also formally endorsed a separate employee letter asking Washington to help pace frontier AI development internationally, aligning with OpenAI and Google DeepMind leadership on that front while standing apart from many of the same companies on open-weight policy. Anthropic is simultaneously the industry's most valuable AI startup and, on China and openness policy specifically, its most isolated major voice.
What to watch: whether Anthropic's more targeted chip-and-distillation policy framework gains traction in Washington relative to the broader open-weight letter, and whether continued criticism from figures like Sacks affects Anthropic's standing with the current administration on other unrelated policy fronts, including its federal AI procurement relationships.
