AI-Native Systems Defined by Autonomous Revision Authority

A recent paper on arXiv offers a clear technical definition for "AI-native" systems, moving past mere marketing jargon. The authors focus on a singular aspect: the authority a system has over its own decisions, particularly revision authority, which refers to the capability of altering system implementations. They differentiate between who makes a decision (occupancy) and revision authority, categorizing the latter into a hierarchy: self-tuning, self-rewriting, and self-architecting. An AI-native system is characterized by an AI's ability to autonomously modify its own implementations. Additionally, the definition includes the necessity for an escalation detector, a verification process, and a verified fallback, while ensuring that purpose and correctness remain under human control. This paper seeks to provide clarity on a term that has previously lacked technical specificity as AI agents increasingly synthesize, verify, and deploy system components.