Exons-Detect: Identifying and Amplifying Exonic Tokens via Hidden-State Discrepancy for Robust AI-Generated Text Detection
Authors: Xiaowei Zhu, Yubing Ren, Fang Fang, Shi Wang, Yanan Cao, Li Guo
Organizations: Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China · School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China · Institute of Computing Science, Chinese Academy of Sciences, Beijing, China
The rapid advancement of large language models has increasingly blurred the boundary between human-written and AI-generated text, raising societal risks such as misinformation dissemination, authorship ambiguity, and threats to intellectual property rights. These concerns highlight the urgent need for effective and reliable detection methods. While existing training-free approaches often achieve strong performance by aggregating token-level signals into a global score, they typically assume uniform token contributions, making them less robust under short sequences or localized token modifications. To address these limitations, we propose Exons-Detect, a training-free method for AI-generated text detection based on an exon-aware token reweighting perspective. Exons-Detect identifies and amplifies informative exonic tokens by measuring hidden-state discrepancy under a dual-model setting, and computes an interpretable translation score from the resulting importance-weighted token sequence. Empirical evaluations demonstrate that Exons-Detect achieves state-of-the-art detection performance and exhibits strong robustness to adversarial attacks and varying input lengths. In particular, it attains a 2.2% relative improvement in average AUROC over the strongest prior baseline on DetectRL.
Figures & tables
Figure 1: Advantages of Our Method Exons-Detect.
Figure 2: Overview of Exons-Detect.
Detectors
M4
DetectRL Multi-LLM
DetectRL Multi-Domain
RealDet
Avg.
AUROC
F1
AUROC
F1
AUROC
F1
AUROC
F1
AUROC
F1
Training-based Methods
OpenAI-D
77.51
71.18
78.15
71.90
74.60
70.03
84.75
77.47
78.75
72.65
Biscope
79.74
73.08
79.97
73.20
76.52
71.64
92.88
86.90
82.28
76.21
R-Detect
61.91
67.14
67.40
66.56
79.19
73.38
65.93
67.72
68.61
68.70
Training-free Methods
Table 1: Detection performance (AUROC and F1 score) on public benchmark datasets.
Figure 3: Detection performance (AUROC curves) against various attacks.
Figure 4: Detection performance on different length.
Figure 5: Detection performance of Exons-Detect under different parameter settings.
Figure 6: Time costs for processing a single sample.
Setting ↓
M4
Multi-L
Multi-D
RealDet
Avg.
Exons-Detect
92.43
90.67
90.46
94.98
92.14
w/o logPPLMW(s^)
91.28
85.32
80.46
93.97
87.76
w/o g(⋅)
91.92
89.63
88.66
94.64
91.21
Model Family ↓
Falcon-7B
92.43
90.67
90.46
94.98
92.14
LLaMA-7B
88.64
92.34
90.31
94.47
91.44
Table 2: Ablation study results under different settings.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Detectors
M4
DetectRL Multi-LLM
DetectRL Multi-Domain
RealDet
Avg.
AUROC
F1
AUROC
F1
AUROC
F1
AUROC
F1
AUROC
F1
Nonlinear mapping
α=10
92.43
88.05
90.67
84.95
90.46
86.59
94.98
91.30
92.14
87.72
α=20
92.25
87.55
90.87
85.20
90.89
87.05
95.02
91.29
92.26
87.77
Linear mapping
α=10
92.24
87.92
90.90
84.93
91.18
87.24
95.03
91.02
92.34
87.78
Appendix
Table 3: Detection performance (AUROC and F1 score) with different mapping functions ( θ=0.15 ).
Parameter Setting
DetectRL Multi-LLM
DetectRL Multi-Domain
Avg.
AUROC
F1
AUROC
F1
AUROC
F1
Reverse
L=2(θ=0.15,α=10)
89.09
84.39
88.24
84.53
88.67
84.46
L=4(θ=0.15,α=10)
89.41
85.07
88.38
84.65
88.90
84.86
L=8(θ=0.15,α=10)
89.13
84.36
88.56
84.70
88.85
84.53
L=16(θ=0.15,α=10)
90.63
85.38
90.55
86.79
90.59
86.09
Appendix
Table 4: Detection performance with different hidden layers.
The growing capability of large language models to produce fluent, contextually coherent text has created mounting pressure on the systems and institutions responsible for ensuring the authenticity of digital content. Advanced generative models such as GPT-4, Claude 3.5, and Llama can produce highly coherent and human-like text, making it increasingly difficult to differentiate between human-written and AI-generated content. While these models have transformative applications, their misuse has raised concerns about misinformation, biased narratives, and security threats. This paper provides a comprehensive analysis of state-of-the-art AI-generated text detection techniques and evaluates their effectiveness through the Counter Turing Test (CT2) shared tasks. Task A (Binary Classification) required participants to distinguish between human-written and AI-generated text, while Task B (Model Attribution) focused on identifying the specific language model responsible for generating a given text. The results demonstrated high performance in binary classification, with the top system achieving an F1 score of 1.0000, but significantly lower scores in model attribution, where the best system achieved 0.9531, highlighting the increased complexity of this task. The top-performing teams leveraged fine-tuned transformer models, ensemble learning, and hybrid detection approaches, with DeBERTa-based and BART-based methods demonstrating strong results. However, the lower scores in Task B underscore the challenges of distinguishing outputs from different LLMs, necessitating further research into adversarial robustness, feature extraction, and cross-domain generalization.
Rajarshi Roy, Gurpreet Singh, Ashhar Aziz +16
Kalyani Government Engineering College, India. · IIIT Guwahati, India. · IIIT Delhi, India. +10
AI-generated text increasingly blends with human writing, raising practical risks such as misinformation, academic misuse, and corpora contamination. While statistical detectors are appealing for efficiency and generalization, they suffer from two key limitations. (i) Boilerplate dominance, boilerplate tokens shared across human and LLM writing can overwhelm discriminative signals. (ii) Brittle point estimates, relying on a single probability score yields unstable decisions under adversarial manipulations. To address these issues, we propose Uncertainty, a multiscale uncertainty estimator that focuses on informative low-probability tokens, which more clearly expose distributional discrepancies. Locally, it alleviates boilerplate dominance by averaging the log-probabilities of low-probability tokens; globally, it reduces brittleness by capturing the distributional shape of this low-probability region via Rényi entropy. We further extend the detector to Uncertainty++ via conditional independent sampling, yielding a more stable uncertainty estimation. Experiments across seven datasets and sixteen LLMs demonstrate high effectiveness, generalization, and robustness. Our code is available at https://github.com/guoyikai2000/Uncertainty-AIGT.
Yikai Guo, Bin Wang, Xilai Fan +2
Beijing Institute of Computer Technology and Application, Beijing, China · Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, China · Southeast University, Nanjing, China +1
The rapid proliferation of Large Language Models has significantly increased the difficulty of distinguishing between human-written and AI generated texts, raising critical issues across academic, editorial, and social domains. This paper investigates the problem of AI generated text detection through the design, implementation, and comparative evaluation of multiple machine learning based detectors. Four neural architectures are developed and analyzed: a Multilayer Perceptron, a one-dimensional Convolutional Neural Network, a MobileNet-based CNN, and a Transformer model. The proposed models are benchmarked against widely used online detectors, including ZeroGPT, GPTZero, QuillBot, Originality.AI, Sapling, IsGen, Rephrase, and Writer. Experiments are conducted on the COLING Multilingual Dataset, considering both English and Italian configurations, as well as on an original thematic dataset focused on Art and Mental Health. Results show that supervised detectors achieve more stable and robust performance than commercial tools across different languages and domains, highlighting key strengths and limitations of current detection strategies.