Detecting Non-Membership in LLM Training Data via Rank Correlations

2026-03-23

AI Summary

This work presents the first systematic approach to non-membership detection in large language model (LLM) training data, a critical capability for copyright enforcement, compliance auditing, and user trust. The proposed method, PRISM, operates under gray-box access—requiring only model logits—and determines whether a given data point was excluded from training by computing the Spearman rank correlation coefficient between its normalized token log-probabilities under the target model and a reference model. Extensive experiments across multiple datasets demonstrate that PRISM reliably identifies non-members with zero false positives, effectively verifying the absence of specific data from an LLM’s training set. These results underscore PRISM’s robustness and practical utility in training data exclusion verification.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

As large language models (LLMs) are trained on increasingly vast and opaque text corpora, determining which data contributed to training has become essential for copyright enforcement, compliance auditing, and user trust. While prior work focuses on detecting whether a dataset was used in training (membership inference), the complementary problem -- verifying that a dataset was not used -- has received little attention. We address this gap by introducing PRISM, a test that detects dataset-level non-membership using only grey-box access to model logits. Our key insight is that two models that have not seen a dataset exhibit higher rank correlation in their normalized token log probabilities than when one model has been trained on that data. Using this observation, we construct a correlation-based test that detects non-membership. Empirically, PRISM reliably rules out membership in training data across all datasets tested while avoiding false positives, thus offering a framework for verifying that specific datasets were excluded from LLM training.

Problem

• non-membership detection
• large language models
• training data verification
• membership inference
• data provenance

Innovation

• non-membership detection
• rank correlation
• large language models
• training data auditing
• grey-box inference