Anthropic CEO Clarifies Stance: No Blanket Ban on Open-Weights AI Models
Dario Amodei outlines targeted safety measures while defending open models without dangerous capabilities as a public good.
Anthropic has never advocated for a ban on open-weights models as a category, CEO Dario Amodei said in a formal statement published on the company's website. The clarification comes as U.S. officials weigh restrictions on Chinese AI models and after a coalition of tech companies signed a letter supporting open-weights access.
A Middle Ground on AI Governance
Amodei's statement positions Anthropic between total openness and protectionist bans. While the company supports open development for models without dangerous capabilities, it calls for mandatory safety testing on all sufficiently powerful AI systems, regardless of whether their weights are publicly released.
"Open-weights models that don't have dangerous capabilities are a public good," Amodei wrote. But he warned that high-capability models pose risks that cannot be retracted once released.
Three Proposed Measures
Anthropic outlined three specific policy recommendations:
1. Ban the sale of powerful chips and AI equipment to China 2. Crack down on industrial-scale model distillation operations 3. Require mandatory safety testing for all sufficiently capable models, open or closed
The company identified two primary "nightmare scenarios" driving its position: authoritarian regimes achieving military superiority through AI, and the misuse of powerful models for cyber or biological attacks.
Industry Divide
The debate has exposed fractures within the AI industry. Nvidia recruited 24 other organizations to defend open-weight AI in a July 2026 letter. OpenAI and Google reportedly signed the coalition letter; Anthropic remained absent.
The tension reflects broader U.S.-China competition over AI supremacy. U.S. officials are considering restrictions on Chinese models to protect national security, while industry players argue that open-weights models are essential for competition and innovation.
What Comes Next
Anthropic's push for mandatory testing suggests a shift toward a regulated licensing or certification model for frontier AI. Such a framework could fundamentally change how open-source AI is developed and distributed globally.
The company's safety-first approach has long distinguished it from competitors. This latest position attempts to balance democratization of AI technology against dual-use risks that emerge at the frontier of capability.
As governments worldwide grapple with AI governance, Anthropic's proposal offers one path forward: preserve openness where it is safe, but impose testing requirements where capabilities cross into dangerous territory.