Anthropic CEO Denies Supporting Ban on Open-Weight AI Models

Following industry criticism, Dario Amodei, CEO of Anthropic, published a blog post denying that his company has ever advocated for a ban on open-weight AI models. The clarification came after Anthropic became the only major frontier AI lab to refuse to sign an open letter, led by Nvidia CEO Jensen Huang, urging U.S. policymakers to avoid premature restrictions on models that allow developers to modify AI on their own infrastructure. The letter was endorsed by over two dozen companies, including OpenAI, Google, Microsoft, and Meta.

While Amodei described open-weight models without dangerous capabilities as a public good that expands competition, he disputed the letter's claim that such models inherently favor defenders over attackers. He cited risks such as the potential to weaponize pandemic-level viruses. Instead of a blanket ban, Amodei proposed three policy measures: tightening export controls on AI chips to China, curbing industrial-scale model distillation, and mandating safety testing for all sufficiently capable models.

The debate coincides with U.S. government concerns over the Chinese startup Moonshot AI. White House and Treasury officials allege Moonshot's Kimi K3 model was trained using distillation of Anthropic's Fable AI model. In response, the Trump administration is considering sanctions and a new independent regulatory framework for AI safety, while Chinese authorities have warned of retaliation against U.S. penalties.