Machine Text Detectors are Membership Inference Attacks
Theoretically and empirically demonstrates the transferability between machine-generated text detection and membership inference attacks, revealing their underlying shared signals in language model distributions.