Karson Elmgren
Senior Researcher, Institute for AI Policy and Strategy (IAPS)
Current frontier AI systems are massive in scale and increasingly autonomous, making traditional human monitoring inadequate to address the governance challenges posed by massive token generation and agentic autonomous actions. In response, a strategy of “using AI to govern AI” may be explored. Through multi-layered technical approaches — including data filtering, character training, mechanistic interpretability research, chain-of-thought monitoring, and AI control research — effective oversight can be achieved across various levels of the AI stack. AI governance requires technically sophisticated institutions capable of fast iteration. Countries such as the United States, the United Kingdom, and China have already explored models including independent nonprofit institutions, specialized research organizations, and other formats for AI governance and verification, forming a diverse governance ecosystem. Nevertheless, governance institutions across countries still need to strengthen coordination and cooperation to collectively address the systemic risks posed by AI and prevent humanity from being overwhelmed by the technological forces it has cultivated.
