Explicit Provenance Needed for Responsible Agentic AI

A recent paper on arXiv (2605.17169) posits that for agentic AI to be deemed responsible, it must have clear provenance throughout its entire lifecycle. The authors argue that existing frameworks do not provide quantifiable, traceable, or actionable provenance, rendering accountability subjective and difficult to enforce when harm results from systems not designed by a single entity. They suggest four areas of focus: demonstrating the essential nature of provenance by pinpointing responsibility gaps in sociotechnical contexts, defining its necessary components via a causal attribution function and responsibility tensor, and promoting computable and actionable accountability.