Retrosynthesis of Synthetic Media for Explainable AI Provenance Forensics

The paper introduces a self‑referential retrosynthesis framework for explainable AI provenance forensics that works with fixed‑generator generative models. It uses a jointly optimized encoder‑decoder pair to embed client inputs, generate high‑fidelity outputs, and then verify provenance by comparing the resynthesized image to the original query. The method eliminates the need for watermarking or generator modifications while providing interpretable evidence of a model’s output origin.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

• rag