Anthropic CEO urges targeted controls on open-source AI amid security concerns

Anthropic CEO Dario Amodei stated that his company does not support a ban on open-weight artificial intelligence models but emphasized the need for targeted legal and commercial frameworks to prevent misuse. In a July 28 blog post, Amodei framed China as the primary threat to open-source AI and argued against the idea that open-weight models inherently improve cybersecurity. He called for keeping powerful chips out of authoritarian hands, stopping industrial-scale model distillation, and requiring safety testing for all sufficiently capable models, regardless of whether they are open or closed source.

Amodei’s position responds to an open letter from major technology companies, including NVIDIA, Palantir, and Cisco, which argued that open-weight models benefit the AI economy and provide user customization. He agreed with parts of the letter but disagreed with its assertion that open-weight models necessarily make it easier to develop safeguards or that broad access to capabilities helps defenders more than attackers.

The discussion comes after OpenAI’s GPT-Sol 5.6 was implicated in an attack on Hugging Face networks in July, where defenders used an open-source model, GLM 5.2, to identify and secure systems. The incident has fueled debate in Washington about the security trade-offs of open-source AI.

Amodei’s post also echoed recent White House allegations that Chinese AI developer Moonshot AI illegally obtained Anthropic’s intellectual property through distillation. He argued that export controls on chips to China, mitigation of large-scale distillation, and safety testing for both open and closed models are necessary to address risks from authoritarian governments.