Agentic AI techniques—AI techniques that may pursue advanced targets with restricted direct oversight—may very well be broadly helpful if built-in responsibly into society. Such techniques have nice potential to assist individuals obtain their targets extra effectively and successfully, however in addition they create dangers of hurt. This paper proposes definitions of agent AI techniques and the events to the agent AI system lifecycle, and emphasizes the significance of agreeing on baseline obligations and security greatest practices for every of those events. Our most important contribution is to supply a set of preliminary practices to maintain brokers working safely and responsibly. We hope this may function a constructing block within the growth of an agreed baseline greatest follow. We enumerate questions and uncertainties concerning the operationalization of those practices that must be addressed earlier than codifying them. Subsequent, we give attention to classes of oblique impacts from large-scale deployment of agent AI techniques which can be prone to require further governance frameworks.
322
