Machine Text Detectors are Membership Inference Attacks

Although membership inference attacks (MIAs) and machine-generated text detection target different goals, their methods often exploit similar signals based on a language model’s probability distribution, and the two tasks have been studied independently. This can result in conclusions that overlook stronger methods and valuable insights from the other task. In this work, we theoretically and empirically demonstrate the transferability, i.e., how well a method originally developed for one task performs on the other, between MIAs and machine text detection. We prove that the metric achieving asymptotically optimal performance is identical for both tasks. We unify existing methods under this optimal metric and hypothesize that the accuracy with which a method approximates this metric is directly correlated with its transferability. Our large-scale empirical experiments demonstrate very strong rank correlation ($ρ\approx 0.7$) in cross-task performance. Notably, we also find that a machine text detector achieves the strongest performance among evaluated methods on both tasks, demonstrating the practical impact of transferability. To facilitate cross-task development and fair evaluation, we introduce MINT, a unified evaluation suite for MIAs and machine-generated text detection, implementing 15 recent methods from both tasks.

💡 Research Summary

The paper “Machine Text Detectors are Membership Inference Attacks” bridges two research strands that have traditionally been treated separately: membership inference attacks (MIAs), which aim to determine whether a given text was part of a language model’s training set, and machine‑generated text detection, which seeks to distinguish human‑written from model‑generated text. Although the goals differ, both tasks rely heavily on the probability that a language model assigns to a piece of text. The authors first formalize each task as a binary hypothesis test and then prove that, under standard asymptotic regularity conditions, the likelihood‑ratio statistic

\