Anthropic's Dario Amodei has clarified his stance on AI model bans, stating that his company does not advocate for such measures. Instead, he is more concerned about the potential misuse of AI by authoritarian governments, particularly China.

In a blog post, Amodei defended open-weight models as a public good, arguing they provide value to businesses and researchers without significant costs. He distinguishes this from his fears regarding Chinese AI labs, which he believes pose a threat due to their growth through potentially unethical means such as stealing intellectual property.

Amodei expresses concerns about the CCP’s capabilities but also notes that other authoritarian regimes could misuse AI for military superiority or repression. His primary worries extend beyond cybersecurity to include the potential for biological attacks, where open-weight models might be more dangerous due to their difficulty in regulation and monitoring once released.

In response, Amodei supports efforts towards creating a global model safety testing organization, hoping for international cooperation to address these concerns. He believes such measures could form a consensus even among adversarial states, hinting at possible collaboration around preventing AI biological weapons.