Kasirzadeh, A., & Gabriel, I. (2025). Characterizing AI Agents for Alignment and Governance. arXiv preprint arXiv:2504.21848.
This paper proposes a framework to understand and govern AI agents based on four key dimensions: autonomy, efficacy, goal complexity, and generality. It argues that effective governance requires understanding these properties and how they vary across different AI agents. The paper provides gradations for each dimension, constructing "agentic profiles" for systems like AlphaGo, ChatGPT, Claude 3.5, and Waymo, to highlight diverse challenges from narrow task-specific assistants to highly autonomous general-purpose systems. It emphasizes the need for governance approaches that align with societal goals, differentiate between risk levels, and adapt to the evolving nature of AI agency.
