Dario Amodei, CEO of Anthropic, has entered the debate surrounding open-weight models, asserting that his company does not support a blanket ban on these models. However, he warned about the risk of authoritarian states gaining access to highly powerful AI systems, revealing a divide within the tech industry between those who see openness as a means to accelerate innovation and those who fear the use of powerful models in cyberattacks or weapon development.

Safe Models are a Public Good

According to a report published by TechCrunch, Amodei stated that Anthropic did not advocate for a ban on open-weight models. He added that models without dangerous capabilities represent a public good for companies, developers, and researchers. His clarification came after a letter signed by NVIDIA, Microsoft, Meta, Hugging Face, and other companies urged governments not to impose broad and early restrictions on open-weight models due to their impact on innovation and competition.

Fear of More Dangerous Capabilities

Amodei focused his concerns on the possibility of powerful governments developing models that grant them permanent military superiority or aid in population surveillance. He also warned against using AI in biological and cyberattacks, suggesting restricting China's access to advanced chips and countering knowledge transfer between models. He supported establishing an international body to test the most capable systems, whether open or closed, with China participating in this system to ensure that global tests are effective and do not become rules applied by one country against its competitors.