Refinement Provenance Inference: Detecting LLM-Refined Training Prompts from Model Behavior

2026-01-05

arXiv.org

AI Summary

This work addresses the challenge of tracing whether prompts used in fine-tuning large language models (LLMs) have been optimized by another LLM—a critical issue for data provenance and dispute resolution in mixed training corpora. The paper formalizes this as the "Rewritten Prompt Inference" (RPI) task and introduces RePro, a novel framework that leverages logit distribution shifts under teacher forcing, integrating both likelihood and ranking signals. By employing shadow fine-tuning to learn transferable representations, RePro enables universal provenance detection without access to original training data, requiring only a lightweight linear head. Experiments demonstrate that RePro consistently achieves strong performance across diverse model architectures and optimizers, effectively capturing intrinsic, optimizer-agnostic distributional shifts rather than superficial stylistic rewriting patterns.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Prompt Engineering / Prompting
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Instruction tuning increasingly relies on LLM-based prompt refinement, where prompts in the training corpus are selectively rewritten by an external refiner to improve clarity and instruction alignment. This motivates an instance-level audit problem: for a fine-tuned model and a training prompt-response pair, can we infer whether the model was trained on the original prompt or its LLM-refined version within a mixed corpus? This matters for dataset governance and dispute resolution when training data are contested. However, it is non-trivial in practice: refined and raw instances are interleaved in the training corpus with unknown, source-dependent mixture ratios, making it harder to develop provenance methods that generalize across models and training setups. In this paper, we formalize this audit task as Refinement Provenance Inference (RPI) and show that prompt refinement yields stable, detectable shifts in teacher-forced token distributions, even when semantic differences are not obvious. Building on this phenomenon, we propose RePro, a logit-based provenance framework that fuses teacher-forced likelihood features with logit-ranking signals. During training, RePro learns a transferable representation via shadow fine-tuning, and uses a lightweight linear head to infer provenance on unseen victims without training-data access. Empirically, RePro consistently attains strong performance and transfers well across refiners, suggesting that it exploits refiner-agnostic distribution shifts rather than rewrite-style artifacts.

Problem

• Refinement Provenance Inference
• LLM-refined prompts
• instruction tuning
• training data provenance
• model auditing

Innovation

• Refinement Provenance Inference
• LLM-refined prompts
• teacher-forced token distribution
• logit-based provenance
• shadow fine-tuning