Anthropic rejects calls to ban open-weight AI

Anthropic CEO Dario Amodei has rejected the idea that the company wants to ban open-weight AI models, while calling for tighter controls on how powerful systems are developed and released.

“Anthropic has never advocated for a ban on open-weights models. Open-weights models that don’t have dangerous capabilities are a public good.”

Amodei published Anthropic’s formal position on Sunday, ending a period of silence that had prompted accusations that the company opposed open weights to protect its business. The statement followed an open letter backed by Nvidia CEO Jensen Huang and dozens of technology companies, as well as reports that US officials were considering a ban on Chinese open-weight models.

Anthropic’s three proposed measures

Rather than banning open weights, Amodei supports three policies:

• Enforce chip export controls and crack down on the “rampant smuggling” of advanced chips into China. Amodei argues that, without US chips, scaling laws prevent China from training models more powerful than American systems.
• Stop industrial-scale distillation, in which labs use outputs from existing models to build capable systems at far lower cost than training from scratch.
• Require safety testing for all sufficiently capable models, whether open or closed and regardless of country of origin. Less capable systems from startups and academic institutions would be exempt entirely.

Amodei agreed with the Nvidia-backed letter that open weights can expand access and competition. He rejected its claim that they necessarily benefit defenders more than attackers, pointing to biological weapons as the clearest example. A sufficiently capable model could potentially help weaponize a pandemic-level virus quickly using widely available materials, while developing defenses against such agents can take years.

Recommended reading

LM Studio Bionic adds Moonshot’s Kimi K3

Distillation is the commercial flashpoint

The position comes after TNW reported that Anthropic’s silence on the open-weights letter had become conspicuous, given the company’s direct commercial exposure to distillation. Anthropic has accused Alibaba’s Qwen lab of conducting its largest-ever distillation campaign against Claude, using 25,000 fake accounts for 29 million exchanges.

Amodei acknowledged that enforcement is “challenging” because accounts may only be identifiable after substantial distillation has already taken place. He called for policy intervention rather than relying on any single company to police the practice.

Anthropic’s distinction is direct: it supports open weights that do not have dangerous capabilities, but opposes models built by extracting knowledge from its closed systems, trained with smuggled chips, or released without safety testing.