Anthropic Denies Seeking Open-Weight AI Ban, Targets China Chip Controls

Anthropic never advocated for a ban on open-weight AI models, CEO Dario Amodei said in a Monday statement, pushing back against rumors the safety-focused lab sought to restrict open-weight competitors.

The clarification follows reports that U.S. officials considered banning Chinese open-weight models, prompting tech companies including Meta, Microsoft, and Nvidia to sign an open letter opposing premature restrictions. Anthropic's absence from that letter fueled speculation the company supported broader bans.

Three-Part Policy Framework

Rather than targeting open-weight distribution, Amodei outlined three specific policy measures: restricting sales of powerful AI chips and equipment to China, cracking down on industrial-scale model distillation, and implementing mandatory safety testing for all capable models, open or closed.

Anthropic views large-scale distillation as a method for Chinese entities to evade chip export controls, building models beyond their hardware's normal limits. The company has raised concerns that distillation campaigns could undermine U.S. semiconductor restrictions designed to limit adversarial AI development.

Capability Over Distribution

Amodei drew a distinction between open-weight models based on their capabilities rather than their licensing structure. "Open-weights models that don't have dangerous capabilities are a public good," he wrote.

However, the CEO argued that open-weight models pose elevated risks for biological and cyberattacks because guardrails are harder to enforce once weights are public and cannot be withdrawn. This framing shifts the regulatory debate from open versus closed sourcing toward capability-based safety thresholds that would apply uniformly across distribution methods.

National Security Focus

The statement redirects attention toward what Amodei described as the primary security threat: authoritarian regimes building superior AI for military use or repression. By emphasizing hardware controls and safety testing over distribution bans, Anthropic is proposing a regulatory framework that targets production means and verified capabilities rather than how models are shared.

The position aligns Anthropic more closely with industry peers who have warned against restrictions that could hamper open-source innovation while still maintaining the company's emphasis on AI safety testing and export controls on advanced computing hardware.