On the Evidentiary Limits of Membership Inference for Copyright Auditing

TL;DR

This paper evaluates the reliability of membership inference attacks for copyright auditing of large language models, demonstrating their fragility against semantic-preserving paraphrasing techniques.

Contribution

It introduces SAGE, a novel paraphrasing framework, and analyzes how it impacts the effectiveness of MIAs in adversarial copyright dispute scenarios.

Findings

MIAs degrade significantly when models are fine-tuned on SAGE paraphrases.

Some leakage persists in certain fine-tuning regimes.

MIAs are brittle and insufficient alone for copyright auditing.

Abstract

As large language models (LLMs) are trained on increasingly opaque corpora, membership inference attacks (MIAs) have been proposed to audit whether copyrighted texts were used during training, despite growing concerns about their reliability under realistic conditions. We ask whether MIAs can serve as admissible evidence in adversarial copyright disputes where an accused model developer may obfuscate training data while preserving semantic content, and formalize this setting through a judge-prosecutor-accused communication protocol. To test robustness under this protocol, we introduce SAGE (Structure-Aware SAE-Guided Extraction), a paraphrasing framework guided by Sparse Autoencoders (SAEs) that rewrites training data to alter lexical structure while preserving semantic content and downstream utility. Our experiments show that state-of-the-art MIAs degrade when models are fine-tuned on SAGE-generated paraphrases, indicating that their signals are not robust to semantics-preserving transformations. While some leakage remains in certain fine-tuning regimes, these results suggest that MIAs are brittle in adversarial settings and insufficient, on their own, as a standalone mechanism for copyright auditing of LLMs.