Anthropic CEO: Open-Weight AI Risks & Safety Tests

Anthropic CEO Dario Amodei maintains his stance that powerful open-weight models pose significant risks. He emphasizes that Anthropic has never called for a ban on these models, stating that less capable ones are a public good. Here's the thing: Amodei worries about two main "nightmare scenarios." First, authoritarian states like China could develop more powerful AI for military or surveillance. Second, open models could be misused for cyber or bio attacks. He notes that AI labs already see such attempts with closed models. What's interesting is Amodei argues open weights are riskier due to a lack of safeguards and the inability to recall released models. He sees a strong imbalance with biological threats, where AI could weaponize viruses much faster than defenses could be built. He is calling for stricter chip export controls to slow China's AI development and wants mandatory safety tests for all powerful models. The bottom line: these discussions highlight ongoing debates about AI safety, national security, and the future of AI development.