ML Security PapersAI-Generated Text is Non-Stationary: Detection via Temporal Tomography
Alva West , Yixuan Weng , Minjun Zhu , Luodan Zhang , Zhen Lin , Guangsheng Bao , Yue Zhang
Published on arXiv
2508.01754
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
TDT achieves 0.855 AUROC on RAID (+7.1%) and +14.1% AUROC on adversarial HART Level 2 paraphrasing tasks by modeling token-level statistical non-stationarity via wavelet decomposition.
Temporal Discrepancy Tomography (TDT)
Novel technique introduced
The field of AI-generated text detection has evolved from supervised classification to zero-shot statistical analysis. However, current approaches share a fundamental limitation: they aggregate token-level measurements into scalar scores, discarding positional information about where anomalies occur. Our empirical analysis reveals that AI-generated text exhibits significant non-stationarity, statistical properties vary by 73.8\% more between text segments compared to human writing. This discovery explains why existing detectors fail against localized adversarial perturbations that exploit this overlooked characteristic. We introduce Temporal Discrepancy Tomography (TDT), a novel detection paradigm that preserves positional information by reformulating detection as a signal processing task. TDT treats token-level discrepancies as a time-series signal and applies Continuous Wavelet Transform to generate a two-dimensional time-scale representation, capturing both the location and linguistic scale of statistical anomalies. On the RAID benchmark, TDT achieves 0.855 AUROC (7.1\% improvement over the best baseline). More importantly, TDT demonstrates robust performance on adversarial tasks, with 14.1\% AUROC improvement on HART Level 2 paraphrasing attacks. Despite its sophisticated analysis, TDT maintains practical efficiency with only 13\% computational overhead. Our work establishes non-stationarity as a fundamental characteristic of AI-generated text and demonstrates that preserving temporal dynamics is essential for robust detection.
Key Contributions
- Empirically establishes that AI-generated text is fundamentally non-stationary, with inter-segment statistical variation 73.8% larger than human text — explaining why scalar-score detectors fail under localized adversarial perturbations.
- Introduces Temporal Discrepancy Tomography (TDT), which applies Continuous Wavelet Transform to token-level discrepancy sequences to produce a 2D time-scale representation capturing anomaly location and linguistic scale.
- Achieves 0.855 AUROC on RAID benchmark (+7.1% over best baseline) and +14.1% AUROC on HART Level 2 paraphrasing attacks with only 13% computational overhead.
🛡️ Threat Analysis
Primary contribution is a novel AI-generated text detection paradigm (TDT) that uses Continuous Wavelet Transform on token-level discrepancies to identify machine-generated content, including robustness against adversarial paraphrasing attacks — squarely output integrity / AI content detection.