Tag
The paper systematically analyzes risks in agentic AI systems induced by expanding cognitive capabilities across physical, social, and self-referential levels, and proposes mitigation strategies to ensure their safe development.