Regulate the deployment, not just the model: a who-does-what plan for AI safety (substack.norabble.com)

Recent discussions in the AI community have shifted focus from mere interest in AI tools and their financial implications to the pressing need for policy and regulation surrounding AI deployment. This shift is driven by growing awareness of the risks associated with AI technologies, necessitating immediate action to establish a structured regulatory framework. The proposed plan advocates for regulating the deployment of AI models rather than just the models themselves, suggesting a multi-layered approach involving trusted deployers, independent evaluators, and a new US-based regulator. Significantly, the plan outlines the creation of a certification system for “trusted deployers,” primarily major cloud providers and AI labs, which would be responsible for implementing robust security measures and compliance standards. These deployers would develop secure environments to host models, ensuring they meet specific safety ratings (AI Deployment Safety Level, or AIDSL). Additionally, the framework includes the establishment of model clearinghouses to manage high-capability models securely. By focusing on the infrastructure and governance of AI deployment, the proposal aims to enhance safety standards and accountability, ensuring that as AI technologies evolve, they are developed and applied within a reliable, ethical, and secure framework.