Artificial intelligence systems are increasingly capable of self-improvement, with some models revising their own outputs, adapting to new data, and even conducting research autonomously. A recent survey of 1,250 arXiv papers reveals a lack of clarity in the terminology used to describe these capabilities, with terms like "self-refine" and "self-evolve" obscuring distinct goals and methods1. This conflation of concepts can hinder progress in the field, as researchers struggle to evaluate and compare different approaches. The development of autonomous research loops, in which AI systems can generate and explore new ideas without human intervention, raises significant implications for the future of AI research. As state-aligned threat activity becomes more prevalent, the potential for AI systems to be used for malicious purposes increases, making it essential for researchers to prioritize clarity and rigor in their work, so what matters most to practitioners is the need to establish a clear and consistent framework for understanding and mitigating the risks associated with recursive self-improvement in AI.