Dario Amodei, CEO of Anthropic, published an essay titled We Must Pace the Frontier on September 12. He argues that the speed of AI development currently exceeds the capacity of existing safety infrastructure. Amodei warns that uncontrolled progress could lead to damages worth hundreds of billions of dollars within 6 to 12 months.
The proposal includes three key pillars for risk management. First, Amodei suggests embedding independent, third-party evaluators with internal access within AI companies. Second, he calls for the adoption of industry-wide safety standards across democratic nations. Third, he advocates for international cooperation on AI governance.
Anthropic has established an internal goal to create tools capable of detecting most model issues by 2027. The essay emphasizes the need for interpretability research to make AI decision-making processes transparent, as current models function as opaque black boxes.
The publication received responses from industry leaders, including Sam Altman of OpenAI, Elon Musk, and Demis Hassabis of Google DeepMind. While they acknowledged the importance of safety, they noted concerns regarding potential impacts on technological leadership and global competition.
