Detecting Non-Membership in LLM Training Data via Rank Correlations

TL;DR

PRISM is a novel method that uses rank correlation of model logits to reliably detect when a dataset was not used in training large language models, aiding in copyright and compliance verification.

Contribution

Introduces PRISM, a correlation-based test for dataset non-membership detection in LLMs using only grey-box model access, addressing a key gap in training data verification.

Findings

PRISM reliably detects non-membership across various datasets.

The method avoids false positives in non-membership detection.

Empirical results show PRISM effectively verifies dataset exclusion.

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.