Ethical perspectives on AI Agents and Agentic AI

Recent developments in Artificial Intelligence (AI) have given rise to AI Agents and Agentic AI. These systems are not only capable of performing narrowly defined tasks, but can also take on more complex activities with minimal to no human involvement. While there is already some technical research on these systems, their ethical evaluation has so far received little attention. This paper provides an initial perspective on how recent developments in AI Agents and Agentic AI relate to classical ethical principles in the context of AI, focusing on transparency, justice and fairness, non-maleficence, responsibility and accountability, and privacy. To enable responsible development and deployment, we highlight three particularly crucial aspects of these systems: (1) their broad agency which (2) has a higher impact on human autonomy, leading to (3) an even deeper entanglement between humans and technology compared to traditional AI systems. Our analysis, which uses the method of the wide reflective equilibrium, serves as a starting point for ethical reflection and underscores the urgent need for ongoing interdisciplinary engagement to ensure that these novel AI systems are developed and used responsibly.