benchmarkscritiqueneutralExisting methods for detecting LLM-generated text mainly focus on document-level classification and cannot identify which parts of the text are generated by LLMsArtificial Intelligence29 Jul 2026http://arxiv.org/abs/2607.21458v1