When AI Settles Down: Late-Stage Stability as a Signature of AI-Generated Text Detection

• URL: http://arxiv.org/abs/2601.04833v1
• Date: Thu, 08 Jan 2026 11:11:00 GMT
• Title: When AI Settles Down: Late-Stage Stability as a Signature of AI-Generated Text Detection
• Abstract summary: We analyze over 120k text samples and reveal Late-Stage Volatility Decay: AI-generated text exhibits rapidly stabilizing log probability fluctuations as generation progresses.<n>We propose two simple features: Derivative Dispersion and Local Volatility, which computed exclusively from late-stage statistics.
• Score: 15.739137990102991
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Zero-shot detection methods for AI-generated text typically aggregate token-level statistics across entire sequences, overlooking the temporal dynamics inherent to autoregressive generation. We analyze over 120k text samples and reveal Late-Stage Volatility Decay: AI-generated text exhibits rapidly stabilizing log probability fluctuations as generation progresses, while human writing maintains higher variability throughout. This divergence peaks in the second half of sequences, where AI-generated text shows 24--32\% lower volatility. Based on this finding, we propose two simple features: Derivative Dispersion and Local Volatility, which computed exclusively from late-stage statistics. Without perturbation sampling or additional model access, our method achieves state-of-the-art performance on EvoBench and MAGE benchmarks and demonstrates strong complementarity with existing global methods.

Related papers

• PURA: Provably Unbiased and Robust Multi-Bit Watermarking for AI-Generated Text Attribution [83.66032773976904]
  We present PURA, a provably unbiased and robust multi-bit watermarking method for text attribution.<n>Instead of perturbing token probabilities directly, PURA embeds payloads in the latent sampling space via keyed inverse transform sampling.<n>We show that PURA substantially outperforms existing unbiased baselines in the high-payload regime.
  arXiv  Detail & Related papers  (2026-08-23T04:44:27Z)
• Hitting a Moving Target: Test-Time Adaptation for AI Text Detection under Continual Distribution Shift [7.332470421231734]
  deployed approaches for AI text detection often rely on training-time access to labeled datasets of both human-written and AI-generated text.<n>We propose a test-time adaptation approach, using semi-supervised learning, that adapts to distribution shifts by leveraging homogeneity among unlabeled samples observed at inference time.<n>We establish that test-time adaptation is a promising framework for AI text detection in the wild.
  arXiv  Detail & Related papers  (2026-06-23T20:37:18Z)
• Cross-Prompt Generalization in Detecting AI-Generated Fake News Using Interpretable Linguistic Features [0.13999481573773068]
  We investigate cross-prompt generalization in fake news detection using three datasets of AI-generated articles produced under distinct prompts.<n>We extract interpretable linguistic features capturing lexical diversity, readability, and emotion-based characteristics.<n>Analysis of feature distributions shows that AI-generated text exhibits increased lexical diversity, reduced readability, and substantially lower emotional intensity compared to the overall dataset.
  arXiv  Detail & Related papers  (2026-06-02T20:33:56Z)
• Breaking the Generator Barrier: Disentangled Representation for Generalizable AI-Text Detection [17.88697261224505]
  Large language models (LLMs) generate text that increasingly resembles human writing.<n>New models emerge rapidly and reduce the robustness of such shortcuts.<n>We propose a progressively structured framework that disentangles AI-detection semantics from generator-aware artifacts.
  arXiv  Detail & Related papers  (2026-04-15T10:14:06Z)
• Human Texts Are Outliers: Detecting LLM-generated Texts via Out-of-distribution Detection [71.59834293521074]
  We develop a framework to distinguish between human-authored and machine-generated text.<n>Our method achieves 98.3% AUROC and AUPR with only 8.9% FPR95 on DeepFake dataset.<n>Code, pretrained weights, and demo will be released.
  arXiv  Detail & Related papers  (2025-10-07T08:14:45Z)
• Temporal-Aware Iterative Speech Model for Dementia Detection [0.0]
  Current methods for automated dementia detection using speech rely on static, time-agnostic features or aggregated linguistic content.<n>We introduce TAI-Speech, a Temporal Aware Iterative framework that dynamically models spontaneous speech for dementia detection.<n>Our work provides a more flexible and robust solution for automated cognitive assessment, operating directly on the dynamics of raw audio.
  arXiv  Detail & Related papers  (2025-09-26T01:56:07Z)
• Diversity Boosts AI-Generated Text Detection [51.56484100374058]
  DivEye is a novel framework that captures how unpredictability fluctuates across a text using surprisal-based features.<n>Our method outperforms existing zero-shot detectors by up to 33.2% and achieves competitive performance with fine-tuned baselines.
  arXiv  Detail & Related papers  (2025-09-23T10:21:22Z)
• Fine-Grained Detection of AI-Generated Text Using Sentence-Level Segmentation [3.088244520495001]
  A sentence-level sequence labeling model proposed to detect transitions between human- and AI-generated text.<n>Our model combines the state-of-the-art pre-trained Transformer models, incorporating Neural Networks (NN) and Conditional Random Fields (CRFs)<n>The evaluation is performed on two publicly available benchmark datasets containing collaborative human and AI-generated texts.
  arXiv  Detail & Related papers  (2025-09-22T14:22:55Z)
• DNA-DetectLLM: Unveiling AI-Generated Text via a DNA-Inspired Mutation-Repair Paradigm [17.258462909671525]
  We introduce DNA-DetectLLM, a zero-shot detection method for distinguishing AI-generated and human-written text.<n>DNA-DetectLLM achieves relative improvements of 5.55% in AUROC and 2.08% in F1 score across multiple public benchmark datasets.
  arXiv  Detail & Related papers  (2025-09-19T03:08:13Z)
• Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Models [57.474294329887236]
  Diffusion large language models (dLLMs) generate text through iterative denoising.<n>Current decoding strategies discard rich intermediate predictions in favor of the final output.<n>We introduce two complementary methods that exploit temporal consistency.
  arXiv  Detail & Related papers  (2025-08-12T17:59:57Z)
• Quality-Aware Language-Conditioned Local Auto-Regressive Anomaly Synthesis and Detection [30.77558600436759]
  ARAS is a language-conditioned, auto-regressive anomaly synthesis approach.<n>It injects local, text-specified defects into normal images via token-anchored latent editing.<n>It significantly enhances defect realism, preserves fine-grained material textures, and provides continuous semantic control over synthesized anomalies.
  arXiv  Detail & Related papers  (2025-08-05T15:07:32Z)
• AI-Generated Text is Non-Stationary: Detection via Temporal Tomography [22.211451973387796]
  We introduce Temporal Discrepancy Tomography (TDT), a novel detection paradigm that preserves positional information by reformulating detection as a signal processing task.<n>On the RAID benchmark, TDT achieves 0.855 AUROC (7.1% improvement over the best baseline)<n>Our work establishes non-stationarity as a fundamental characteristic of AI-generated text and demonstrates that preserving temporal dynamics is essential for robust detection.
  arXiv  Detail & Related papers  (2025-08-03T13:43:34Z)
• Group-Adaptive Threshold Optimization for Robust AI-Generated Text Detection [58.419940585826744]
  We introduce FairOPT, an algorithm for group-specific threshold optimization for probabilistic AI-text detectors.<n>We partitioned data into subgroups based on attributes (e.g., text length and writing style) and implemented FairOPT to learn decision thresholds for each group to reduce discrepancy.<n>Our framework paves the way for more robust classification in AI-generated content detection via post-processing.
  arXiv  Detail & Related papers  (2025-02-06T21:58:48Z)
• Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing [71.29488677105127]
  Existing scene text recognition (STR) methods struggle to recognize challenging texts, especially for artistic and severely distorted characters. We propose a contrastive learning-based STR framework by leveraging synthetic and real unlabeled data without any human cost. Our method achieves SOTA performance (94.7% and 70.9% average accuracy on common benchmarks and Union14M-Benchmark.
  arXiv  Detail & Related papers  (2024-11-23T15:24:47Z)
• Detecting Machine-Generated Long-Form Content with Latent-Space Variables [54.07946647012579]
  Existing zero-shot detectors primarily focus on token-level distributions, which are vulnerable to real-world domain shifts. We propose a more robust method that incorporates abstract elements, such as event transitions, as key deciding factors to detect machine versus human texts.
  arXiv  Detail & Related papers  (2024-10-04T18:42:09Z)
• Improving Sampling Methods for Fine-tuning SentenceBERT in Text Streams [49.3179290313959]
  This study explores the efficacy of seven text sampling methods designed to selectively fine-tune language models. We precisely assess the impact of these methods on fine-tuning the SBERT model using four different loss functions. Our findings indicate that Softmax loss and Batch All Triplets loss are particularly effective for text stream classification.
  arXiv  Detail & Related papers  (2024-03-18T23:41:52Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.